Akshay 🚀's banner

Akshay 🚀

@akshay_pachaar • 285,521 subscribers

Simplifying LLMs, AI Agents, RAG, and Machine Learning for you! • Co-founder @dailydoseofds_• BITS Pilani • 3 Patents • ex-AI Engineer @ LightningAI

Shorts

Don't train the model, evolve the harness. I read a brilliant blog post from Hugging Face where they took a frozen open model scoring 0% on a hard legal agent benchmark, left its weights alone, and let an automated loop rewrite only the code around it. That code layer is the harness, the runtime wrapper that feeds the model context, runs its tool calls, and decides when a run ends. By the time the loop finished, the system had essentially matched Sonnet 4.6 on the benchmark's headline metric, at roughly 7x lower cost per task. Zero weights changed. The gain existed because of where the model was failing. The judge only grades files saved in the right place under the exact requested filename, and the model kept doing the legal analysis correctly, then saving it under the wrong name, dropping it in a scratch folder, or never writing it at all. So the 0% was never measuring legal reasoning. It was measuring the harness. Hand-tuning that layer is slow and model-specific, so they automated it. A Claude proposer adds exactly one mechanism per iteration, and an outer loop keeps it only if it clearly beats the current best, so accepted mechanisms compound. What the loop discovered says a lot about where agents actually fail. → The biggest single gain was file handling, not intelligence. An automatic step that lands the deliverable exactly where the judge expects it beat every prompt change, with zero extra model tokens. → Code fixes transferred across models, prompt playbooks did not. The same harness lifted a smaller model from the same family by 14 points, but the tuned prompts hurt a different model family on tasks it could already finish. → The harness mattered more than anything else. Same model, same judge, same tasks, and five different harnesses scored anywhere between 3.5% and 80.1%. The gains do eventually flatten, and the remaining misses look like real capability gaps. At some point the wrapper runs out of tricks and the model has to carry the work. But the lesson holds. A benchmark score measures the model and its harness together, and until the harness is fixed, it's impossible to know which one failed. I highly recommend reading this: I also wrote a deep dive on agent harness engineering a while back, covering the orchestration loop, tools, memory, context management, and everything that turns a stateless LLM into a capable agent. The article is quoted below.

Don't train the model, evolve the harness. I read a brilliant blog post from Hugging Face where they took a frozen open model scoring 0% on a hard legal agent benchmark, left its weights alone, and let an automated loop rewrite only the code around it. That code layer is the harness, the runtime wrapper that feeds the model context, runs its tool calls, and decides when a run ends. By the time the loop finished, the system had essentially matched Sonnet 4.6 on the benchmark's headline metric, at roughly 7x lower cost per task. Zero weights changed. The gain existed because of where the model was failing. The judge only grades files saved in the right place under the exact requested filename, and the model kept doing the legal analysis correctly, then saving it under the wrong name, dropping it in a scratch folder, or never writing it at all. So the 0% was never measuring legal reasoning. It was measuring the harness. Hand-tuning that layer is slow and model-specific, so they automated it. A Claude proposer adds exactly one mechanism per iteration, and an outer loop keeps it only if it clearly beats the current best, so accepted mechanisms compound. What the loop discovered says a lot about where agents actually fail. → The biggest single gain was file handling, not intelligence. An automatic step that lands the deliverable exactly where the judge expects it beat every prompt change, with zero extra model tokens. → Code fixes transferred across models, prompt playbooks did not. The same harness lifted a smaller model from the same family by 14 points, but the tuned prompts hurt a different model family on tasks it could already finish. → The harness mattered more than anything else. Same model, same judge, same tasks, and five different harnesses scored anywhere between 3.5% and 80.1%. The gains do eventually flatten, and the remaining misses look like real capability gaps. At some point the wrapper runs out of tricks and the model has to carry the work. But the lesson holds. A benchmark score measures the model and its harness together, and until the harness is fixed, it's impossible to know which one failed. I highly recommend reading this: I also wrote a deep dive on agent harness engineering a while back, covering the orchestration loop, tools, memory, context management, and everything that turns a stateless LLM into a capable agent. The article is quoted below.

244,990 views

Karpathy's Agentic Engineering finally has proper tooling! (built by Google) Karpathy defined agentic engineering as the discipline that separates production agent work from vibe coding. The core skills he listed were spec design, eval loops, and security oversight. The problem has been that practicing this still requires a different tool for every phase: - editor for code - a terminal for scaffolding - a browser for testing - a cloud console for deployment - and a separate framework for evals. Every transition is a context switch. The solution to production-grade Agentic Engineering is now actually implemented in Google’s Agents CLI. It covers the entire workflow in one place for scaffolding, evaluating, and deploying ADK agents. One setup command injects 7 ADK-specific skills into a coding agent's context, which lets it handle scaffolding, evals, deployment, and enterprise registration through natural language. I tested this end-to-end by building a RAG agent from scratch using Claude Code. It scaffolded the full project from the ADK agentic_rag template, generated 20 eval scenarios with LLM-as-judge scoring, and returned a quantitative scorecard. Finally, it also deployed everything to Agent Runtime and registered the agent to Gemini Enterprise, so the entire org can discover and use it. The video below shows this in action, and I worked with the Google Cloud team to put this together. Agents CLI GitHub repo → (don't forget to star it ⭐ ) I wrote up the full build covering all six steps from install to enterprise registration. It includes the eval scorecard, the instruction loophole the eval caught before deployment, and what the deployment process actually looks like end-to-end. Read it below.

Karpathy's Agentic Engineering finally has proper tooling! (built by Google) Karpathy defined agentic engineering as the discipline that separates production agent work from vibe coding. The core skills he listed were spec design, eval loops, and security oversight. The problem has been that practicing this still requires a different tool for every phase: - editor for code - a terminal for scaffolding - a browser for testing - a cloud console for deployment - and a separate framework for evals. Every transition is a context switch. The solution to production-grade Agentic Engineering is now actually implemented in Google’s Agents CLI. It covers the entire workflow in one place for scaffolding, evaluating, and deploying ADK agents. One setup command injects 7 ADK-specific skills into a coding agent's context, which lets it handle scaffolding, evals, deployment, and enterprise registration through natural language. I tested this end-to-end by building a RAG agent from scratch using Claude Code. It scaffolded the full project from the ADK agentic_rag template, generated 20 eval scenarios with LLM-as-judge scoring, and returned a quantitative scorecard. Finally, it also deployed everything to Agent Runtime and registered the agent to Gemini Enterprise, so the entire org can discover and use it. The video below shows this in action, and I worked with the Google Cloud team to put this together. Agents CLI GitHub repo → (don't forget to star it ⭐ ) I wrote up the full build covering all six steps from install to enterprise registration. It includes the eval scorecard, the instruction loophole the eval caught before deployment, and what the deployment process actually looks like end-to-end. Read it below.

258,436 views

Real-time object detection will never be the same. Traditional YOLO needs NMS to remove duplicate boxes; it's slow and inconsistent. YOLO26 skips it entirely: single-pass predictions, faster inference and up to 300 detections per image. Download model:

Real-time object detection will never be the same. Traditional YOLO needs NMS to remove duplicate boxes; it's slow and inconsistent. YOLO26 skips it entirely: single-pass predictions, faster inference and up to 300 detections per image. Download model:

374,302 views

50% cheaper Claude inference with just one line of code change! - Remove → model="claude-opus-4-8" - Add → model="ship-like/claude-opus-4-8" I verified the cost saving in my own terminal by invoking the same Anthropic model with the same prompt. The underlying engineering by Ship is actually interesting, and the patterns can be used in any production LLM stack. Essentially, a trained model is a frozen artifact. Every request performs the same forward-pass, whether it extracts a date or refactors a module, because the compute decision was made at training time, before the request existed. Ship makes that decision at inference time instead. After seeing a request, it searches over executions, involving single models, cascades, ensembles, or harnesses with tools, and serves the cheapest one that will match the reference model's quality. This is not a basic router, because picking a cheaper model per query doesn't ensure the cheaper model preserves the original's behavior, like output shape, tool-call patterns, and refusals. Ship measures this equivalence directly. Outputs stay distributionally indistinguishable from the reference model, not token-identical, since two calls to the same model already differ, but they are indistinguishable in capability and behavior. Of course, some requests execute cheaply and some cost Ship more than the customer pays, but the price per request is still a flat 50% off either way, so the execution-cost variance moves off the application's bill entirely. The video below depicts the cost savings and output in my real invocation, and I partnered with the team to put this together.

50% cheaper Claude inference with just one line of code change! - Remove → model="claude-opus-4-8" - Add → model="ship-like/claude-opus-4-8" I verified the cost saving in my own terminal by invoking the same Anthropic model with the same prompt. The underlying engineering by Ship is actually interesting, and the patterns can be used in any production LLM stack. Essentially, a trained model is a frozen artifact. Every request performs the same forward-pass, whether it extracts a date or refactors a module, because the compute decision was made at training time, before the request existed. Ship makes that decision at inference time instead. After seeing a request, it searches over executions, involving single models, cascades, ensembles, or harnesses with tools, and serves the cheapest one that will match the reference model's quality. This is not a basic router, because picking a cheaper model per query doesn't ensure the cheaper model preserves the original's behavior, like output shape, tool-call patterns, and refusals. Ship measures this equivalence directly. Outputs stay distributionally indistinguishable from the reference model, not token-identical, since two calls to the same model already differ, but they are indistinguishable in capability and behavior. Of course, some requests execute cheaply and some cost Ship more than the customer pays, but the price per request is still a flat 50% off either way, so the execution-cost variance moves off the application's bill entirely. The video below depicts the cost savings and output in my real invocation, and I partnered with the team to put this together.

63,725 views

Hermes meets SuperGrok! xAI just made every SuperGrok subscription work inside Hermes Agent. One browser login, no API key, no separate billing. And it doesn't just unlock text chat with Grok 4.3. The same OAuth token gives the agent access to: → Grok Text-to-Speech for spoken responses → Grok Imagine for image and video generation → x_search for real-time X/Twitter search I just added a new X Research Agent profile to my Hermes. Now my agent watches X while I ship. Setup takes about 60 seconds: Available on every SuperGrok tier, no restrictions. I wrote a full deep dive covering Hermes agent's architecture, memory system, self-evolving skills, GEPA optimization, and setting up multiple specialized agents The article is quoted below.

147,042 views

Big moment for Postgres! Search has always been Postgres' weak spot, and everyone just accepted it. If you needed a real relevance-ranked keyword search, the default answer was to spin up Elasticsearch or add Algolia and deal with the data sync headaches forever. The problem isn't that Postgres can't do text search. It can. But the built-in `ts_rank` function uses a basic term frequency algorithm that doesn't come close to what modern search engines deliver. So teams end up: - Running a separate Elasticsearch cluster just for search - Building sync pipelines that inevitably drift out of consistency - Paying for managed search services that charge per query - Accepting mediocre search relevance because "good enough" ships faster But this is actually a solvable problem. You can realistically bring industry-standard search ranking directly into Postgres, which eliminates the need for external infra entirely. This exact solution is now available with the newly open-sourced pg_textsearch by Tiger Data - Creators of TimescaleDB, a Postgres extension that brings true BM25 relevance ranking into the database. BM25 is the algorithm behind Elasticsearch, Lucene, and most modern search engines. Now it runs natively in Postgres. Here's what pg_textsearch enables: - True BM25 ranking with configurable parameters (the same algorithm powering production search systems) - Simple SQL syntax: `ORDER BY content 'search terms'` - Works with Postgres text search configurations for multiple languages - Pairs naturally with pgvector for hybrid keyword + semantic search That last point matters a lot for RAG apps. The video below shows this in action, and I worked with the team to put this together. You can now do hybrid retrieval (combining keyword matching with vector similarity) in a single database, without stitching together multiple systems. The syntax is clean enough that you can add relevance-ranked search to existing queries in minutes. pg_textsearch is fully open-source under the PostgreSQL license. You can find a link to their GitHub repo in the next tweet.

Big moment for Postgres! Search has always been Postgres' weak spot, and everyone just accepted it. If you needed a real relevance-ranked keyword search, the default answer was to spin up Elasticsearch or add Algolia and deal with the data sync headaches forever. The problem isn't that Postgres can't do text search. It can. But the built-in `ts_rank` function uses a basic term frequency algorithm that doesn't come close to what modern search engines deliver. So teams end up: - Running a separate Elasticsearch cluster just for search - Building sync pipelines that inevitably drift out of consistency - Paying for managed search services that charge per query - Accepting mediocre search relevance because "good enough" ships faster But this is actually a solvable problem. You can realistically bring industry-standard search ranking directly into Postgres, which eliminates the need for external infra entirely. This exact solution is now available with the newly open-sourced pg_textsearch by Tiger Data - Creators of TimescaleDB, a Postgres extension that brings true BM25 relevance ranking into the database. BM25 is the algorithm behind Elasticsearch, Lucene, and most modern search engines. Now it runs natively in Postgres. Here's what pg_textsearch enables: - True BM25 ranking with configurable parameters (the same algorithm powering production search systems) - Simple SQL syntax: `ORDER BY content 'search terms'` - Works with Postgres text search configurations for multiple languages - Pairs naturally with pgvector for hybrid keyword + semantic search That last point matters a lot for RAG apps. The video below shows this in action, and I worked with the team to put this together. You can now do hybrid retrieval (combining keyword matching with vector similarity) in a single database, without stitching together multiple systems. The syntax is clean enough that you can add relevance-ranked search to existing queries in minutes. pg_textsearch is fully open-source under the PostgreSQL license. You can find a link to their GitHub repo in the next tweet.

215,667 views

Neural network visualized!

Neural network visualized!

571,666 views

Drag-and-drop UI to build AI agents! Langflow is a powerful visual tool for building and deploying AI-powered agents and workflows—without writing any code. Supports all major LLMs, vector DBs, etc. 100% open-source, 62k+ stars 🌟

Drag-and-drop UI to build AI agents! Langflow is a powerful visual tool for building and deploying AI-powered agents and workflows—without writing any code. Supports all major LLMs, vector DBs, etc. 100% open-source, 62k+ stars 🌟

208,002 views

AI agents can finally talk to your frontend! The AG-UI Protocol bridges the critical gap between AI agents and frontend apps, making human-agent collaboration seamless. MCP: Agents to tools A2A: Agents to agents AG-UI: Agents to users 100% open-source.

AI agents can finally talk to your frontend! The AG-UI Protocol bridges the critical gap between AI agents and frontend apps, making human-agent collaboration seamless. MCP: Agents to tools A2A: Agents to agents AG-UI: Agents to users 100% open-source.

188,893 views

Fine-tune DeepSeek-OCR on your own language! (100% local) DeepSeek-OCR is a 3B-parameter vision model that achieves 97% precision while using 10× fewer vision tokens than text-based LLMs. It handles tables, papers, and handwriting without killing your GPU or budget. Why it matters: Most vision models treat documents as massive sequences of tokens, making long-context processing expensive and slow. DeepSeek-OCR uses context optical compression to convert 2D layouts into vision tokens, enabling efficient processing of complex documents. The best part? You can easily fine-tune it for your specific use case on a single GPU. I used Unsloth to run this experiment on Persian text and saw an 88.26% improvement in character error rate. ↳ Base model: 149% character error rate (CER) ↳ Fine-tuned model: 60% CER (57% more accurate) ↳ Training time: 60 steps on a single GPU Persian was just the test case. You can swap in your own dataset for any language, document type, or specific domain you're working with. I've shared the complete guide in the next tweet - all the code, notebooks, and environment setup ready to run with a single click. Everything is 100% open-source!

Fine-tune DeepSeek-OCR on your own language! (100% local) DeepSeek-OCR is a 3B-parameter vision model that achieves 97% precision while using 10× fewer vision tokens than text-based LLMs. It handles tables, papers, and handwriting without killing your GPU or budget. Why it matters: Most vision models treat documents as massive sequences of tokens, making long-context processing expensive and slow. DeepSeek-OCR uses context optical compression to convert 2D layouts into vision tokens, enabling efficient processing of complex documents. The best part? You can easily fine-tune it for your specific use case on a single GPU. I used Unsloth to run this experiment on Persian text and saw an 88.26% improvement in character error rate. ↳ Base model: 149% character error rate (CER) ↳ Fine-tuned model: 60% CER (57% more accurate) ↳ Training time: 60 steps on a single GPU Persian was just the test case. You can swap in your own dataset for any language, document type, or specific domain you're working with. I've shared the complete guide in the next tweet - all the code, notebooks, and environment setup ready to run with a single click. Everything is 100% open-source!

126,213 views

Google released Gemma 3 270M, a new model for hyper-efficient local AI! We'll fine-tune this model and make it very smart at playing chess and predict the next move. Tech stack: - Unsloth AI for efficient fine-tuning. - Hugging Face transformers to run it locally. Let's go! 🚀

Google released Gemma 3 270M, a new model for hyper-efficient local AI! We'll fine-tune this model and make it very smart at playing chess and predict the next move. Tech stack: - Unsloth AI for efficient fine-tuning. - Hugging Face transformers to run it locally. Let's go! 🚀

145,802 views

Google just open-sourced LangExtract Python library! It uses LLMs to extract entities, attributes, and relations—with exact source grounding—from unstructured documents. Flexible LLM support (Gemini, OpenAI, Ollama) 100% open-source.

Google just open-sourced LangExtract Python library! It uses LLMs to extract entities, attributes, and relations—with exact source grounding—from unstructured documents. Flexible LLM support (Gemini, OpenAI, Ollama) 100% open-source.

120,405 views

OpenClaw, but built for normal people. Sim is an open-source platform that lets you build AI agent workflows on a drag-and-drop canvas. Connect them to channels like Telegram and WhatsApp and deploy without writing a single line of code. They also have a built-in Copilot that generates entire workflows from plain English, which you can then tweak and customize in the UI. Key features: - Free and open-source (Apache 2.0) - Vector store integration for RAG-grounded agents - Self-host with one command (`npx simstudio`) - Run fully local with Ollama, no API keys needed - Supports vLLM for production-grade self-hosted inference The thing I really like about Sim is the level of control you get. You can add conditional branching, parallel execution, human-in-the-loop approval gates, and even nest workflows inside other workflows. Everything is visible on the canvas, so you know exactly what your agent is doing at every step. And you can build a workflow in Sim, deploy it as an MCP server, and plug it into any agent, including OpenClaw. I've shared the link to Sim's GitHub repo in the next tweet.

OpenClaw, but built for normal people. Sim is an open-source platform that lets you build AI agent workflows on a drag-and-drop canvas. Connect them to channels like Telegram and WhatsApp and deploy without writing a single line of code. They also have a built-in Copilot that generates entire workflows from plain English, which you can then tweak and customize in the UI. Key features: - Free and open-source (Apache 2.0) - Vector store integration for RAG-grounded agents - Self-host with one command (`npx simstudio`) - Run fully local with Ollama, no API keys needed - Supports vLLM for production-grade self-hosted inference The thing I really like about Sim is the level of control you get. You can add conditional branching, parallel execution, human-in-the-loop approval gates, and even nest workflows inside other workflows. Everything is visible on the canvas, so you know exactly what your agent is doing at every step. And you can build a workflow in Sim, deploy it as an MCP server, and plug it into any agent, including OpenClaw. I've shared the link to Sim's GitHub repo in the next tweet.

52,426 views

Transformers & LLMs cheatsheets for Stanford's CME-295! Covering tokenization, self-attention, prompting, fine-tuning, LLM-as-a-judge, RAG, AI Agents, and reasoning models. 100% free and open-source.

Transformers & LLMs cheatsheets for Stanford's CME-295! Covering tokenization, self-attention, prompting, fine-tuning, LLM-as-a-judge, RAG, AI Agents, and reasoning models. 100% free and open-source.

101,649 views

Turn complex docs into clean, LLM-ready data! Every AI company I've talked to is solving the same problem: how do you build systems that don't hallucinate and back up every answer with proper citations? Tensorlake is a tool that extracts custom-defined structured data from any unstructured document in 3 steps: ↳ Define your schema ↳ Enable citations ↳ Extract You get RAG-ready data with precise citations and bounding boxes. Feed this to your LLM, and you'll generate responses that are citation-backed and fully auditable. This is the difference between a demo and a production system. When your AI can show exactly where it got its information, you move from proof-of-concept to something people can actually trust and deploy. I've shared the Tensorlake GitHub repo in the replies!

Turn complex docs into clean, LLM-ready data! Every AI company I've talked to is solving the same problem: how do you build systems that don't hallucinate and back up every answer with proper citations? Tensorlake is a tool that extracts custom-defined structured data from any unstructured document in 3 steps: ↳ Define your schema ↳ Enable citations ↳ Extract You get RAG-ready data with precise citations and bounding boxes. Feed this to your LLM, and you'll generate responses that are citation-backed and fully auditable. This is the difference between a demo and a production system. When your AI can show exactly where it got its information, you move from proof-of-concept to something people can actually trust and deploy. I've shared the Tensorlake GitHub repo in the replies!

58,172 views

K-Means is simple. Making it fast on GPU isn't. Flash-KMeans is an IO-aware implementation of exact k-means that rethinks the algorithm around modern GPU bottlenecks. By attacking the memory bottlenecks directly, Flash-KMeans achieves: - 30x speedup over cuML - 200x speedup over FAISS Using the same exact algorithm, just engineered for today’s hardware. At the million-scale, Flash-KMeans can complete a k-means iteration in milliseconds. Here's why this matters today: K-means has always been an offline primitive. Something you run once to preprocess data and move on. These speedups change that. ↳ Vector databases like FAISS use k-means to build search indices. Faster k-means means you can re-index dynamically as data changes, not batch it overnight. ↳ LLM quantization methods need k-means to find optimal weight codebooks, per layer, repeatedly. What takes hours could now take minutes. ↳ MoE models need fast token routing at inference time. Millisecond k-means makes it viable to run this inside the inference loop, not just in preprocessing. The 200x over FAISS is the number to internalize. FAISS is the industry standard. Most production vector search systems sit on top of it. Link to the paper and code in next tweet!

K-Means is simple. Making it fast on GPU isn't. Flash-KMeans is an IO-aware implementation of exact k-means that rethinks the algorithm around modern GPU bottlenecks. By attacking the memory bottlenecks directly, Flash-KMeans achieves: - 30x speedup over cuML - 200x speedup over FAISS Using the same exact algorithm, just engineered for today’s hardware. At the million-scale, Flash-KMeans can complete a k-means iteration in milliseconds. Here's why this matters today: K-means has always been an offline primitive. Something you run once to preprocess data and move on. These speedups change that. ↳ Vector databases like FAISS use k-means to build search indices. Faster k-means means you can re-index dynamically as data changes, not batch it overnight. ↳ LLM quantization methods need k-means to find optimal weight codebooks, per layer, repeatedly. What takes hours could now take minutes. ↳ MoE models need fast token routing at inference time. Millisecond k-means makes it viable to run this inside the inference loop, not just in preprocessing. The 200x over FAISS is the number to internalize. FAISS is the industry standard. Most production vector search systems sit on top of it. Link to the paper and code in next tweet!

36,317 views

Check this!! Now you can scrape ANY website by just writing a prompt. Using 's /extract endpoint, just describe what you want to extract in a prompt. This produces LLM-ready structured output. No more hard coding!

Check this!! Now you can scrape ANY website by just writing a prompt. Using 's /extract endpoint, just describe what you want to extract in a prompt. This produces LLM-ready structured output. No more hard coding!

64,158 views

Big moment for text-to-speech. Qwen just open-sourced a text-to-speech model that lets you clone voices, design new ones, and control speech using natural language. Let me explain what I mean: You can literally tell it "speak in a cheerful tone with slight nervousness," and it actually does that. No complex audio engineering needed. What makes this special: - 3-second voice cloning - Covers 10 languages: English, German, French, and more - Latency as low as 97ms for real-time applications - Supports both streaming and non-streaming generation The model comes in two sizes (0.6B and 1.7B parameters), so you can pick based on your hardware and quality needs. Three modes to work with: 1. Custom Voice: Use pre-built premium voices with instruction-based style control 2. Voice Design: Describe the voice you want in plain English (or Chinese), and the model creates it 3. Voice Clone: Provide a 3-second reference audio and clone that voice The best part? It integrates with vLLM for production deployment and has a simple Python package you can pip install. I've shared a link to the GitHub repo in the next tweet.

Big moment for text-to-speech. Qwen just open-sourced a text-to-speech model that lets you clone voices, design new ones, and control speech using natural language. Let me explain what I mean: You can literally tell it "speak in a cheerful tone with slight nervousness," and it actually does that. No complex audio engineering needed. What makes this special: - 3-second voice cloning - Covers 10 languages: English, German, French, and more - Latency as low as 97ms for real-time applications - Supports both streaming and non-streaming generation The model comes in two sizes (0.6B and 1.7B parameters), so you can pick based on your hardware and quality needs. Three modes to work with: 1. Custom Voice: Use pre-built premium voices with instruction-based style control 2. Voice Design: Describe the voice you want in plain English (or Chinese), and the model creates it 3. Voice Clone: Provide a 3-second reference audio and clone that voice The best part? It integrates with vLLM for production deployment and has a simple Python package you can pip install. I've shared a link to the GitHub repo in the next tweet.

31,249 views

Videos

Anya Rossi

sweetdream.ai

SweetDream.ai•Sponsored•Livecam

Watch Anya Live

Anya is streaming live right now! Join her private show and enjoy exclusive content.

Exclusive private shows

1.2k viewers online

Private Show

Join now for exclusive access

Free preview available • Premium content

UC Berkeley just open-sourced FreeToken. (2–4x faster local LLM inference than Ollama) the results are wild: - Qwen3.6-35B on an 8GB GPU at 39.3 tokens/s - DeepSeek-V4-Flash 284B on a 32GB GPU at 22 tokens/s - GLM-5.2 753B on a 96GB GPU at 14.9 tokens/s a 35B model at 16-bit precision needs about 70GB just for its weights. even at 4 bits it is close to 18GB, and FreeToken serves it on an 8GB GPU. let me explain how: all three models mentioned above are Mixture-of-Experts, and that is what FreeToken takes advantage of. each layer holds hundreds of separate experts plus a small router that picks a few of them per token. Qwen3.6-35B activates roughly 3B of its 35B parameters per token. DeepSeek-V4-Flash picks 6 of 256 experts per layer, so 13B of its 284B run at a time. so compute was never the bottleneck. the weights a single step touches fit comfortably on a consumer GPU. every expert the router might pick still has to exist somewhere. they sit in system RAM, and the GPU keeps a cache of the ones the model has been using recently. so everything comes down to what happens when the router picks an expert that is not on the GPU. there are two ways to serve that miss: 1. copy it over PCIe and run it on the GPU 2. run it on the CPU, where it already lives both read from the same system memory, so they compete for one pool of bandwidth instead of adding to each other. existing engines pick one option and freeze it when the model loads. but routing changes on every token, so a fixed choice misses most of what the model asks for. FreeToken measures both bandwidths on your machine and splits each step's misses between the two paths in proportion. the GPU and CPU results then merge exactly, with no approximation. two machines with the same GPU can end up wanting opposite strategies, which I did not expect. a 5090 in a gaming desktop should push nearly everything over PCIe, while an 8GB laptop is better off computing most misses on the CPU. none of that is readable off a spec sheet, so the engine profiles it once per machine. the second half of the design is about agents. coding agents constantly rewrite their own history, and every edit normally forces thousands of tokens back through prefill. FreeToken saves its checkpoints at the exact boundaries agent frameworks cut on, so it only reprocesses the new part. its slowest first token stays under 44 seconds, while llama.cpp peaks at 232 and KTransformers at 946. it serves the OpenAI and Anthropic APIs under Apache 2.0, so Claude Code and Codex can point at it directly. releasing weights publicly decides who can download a model, not who can afford to run one. frontier open models keep shipping, and running them still assumes a rented cluster. meanwhile there are over a hundred million consumer machines with discrete GPUs sitting mostly idle. closing that gap was never a hardware problem, and work like this is what turns open weights into something you can actually use. paper: repo: almost every idea in this post, from why memory bandwidth decides the outcome to why moving weights costs more than computing on them, comes straight out of how a GPU is built. I wrote a detailed primer on that. the article is quoted below.

UC Berkeley just open-sourced FreeToken. (2–4x faster local LLM inference than Ollama) the results are wild: - Qwen3.6-35B on an 8GB GPU at 39.3 tokens/s - DeepSeek-V4-Flash 284B on a 32GB GPU at 22 tokens/s - GLM-5.2 753B on a 96GB GPU at 14.9 tokens/s a 35B model at 16-bit precision needs about 70GB just for its weights. even at 4 bits it is close to 18GB, and FreeToken serves it on an 8GB GPU. let me explain how: all three models mentioned above are Mixture-of-Experts, and that is what FreeToken takes advantage of. each layer holds hundreds of separate experts plus a small router that picks a few of them per token. Qwen3.6-35B activates roughly 3B of its 35B parameters per token. DeepSeek-V4-Flash picks 6 of 256 experts per layer, so 13B of its 284B run at a time. so compute was never the bottleneck. the weights a single step touches fit comfortably on a consumer GPU. every expert the router might pick still has to exist somewhere. they sit in system RAM, and the GPU keeps a cache of the ones the model has been using recently. so everything comes down to what happens when the router picks an expert that is not on the GPU. there are two ways to serve that miss: 1. copy it over PCIe and run it on the GPU 2. run it on the CPU, where it already lives both read from the same system memory, so they compete for one pool of bandwidth instead of adding to each other. existing engines pick one option and freeze it when the model loads. but routing changes on every token, so a fixed choice misses most of what the model asks for. FreeToken measures both bandwidths on your machine and splits each step's misses between the two paths in proportion. the GPU and CPU results then merge exactly, with no approximation. two machines with the same GPU can end up wanting opposite strategies, which I did not expect. a 5090 in a gaming desktop should push nearly everything over PCIe, while an 8GB laptop is better off computing most misses on the CPU. none of that is readable off a spec sheet, so the engine profiles it once per machine. the second half of the design is about agents. coding agents constantly rewrite their own history, and every edit normally forces thousands of tokens back through prefill. FreeToken saves its checkpoints at the exact boundaries agent frameworks cut on, so it only reprocesses the new part. its slowest first token stays under 44 seconds, while llama.cpp peaks at 232 and KTransformers at 946. it serves the OpenAI and Anthropic APIs under Apache 2.0, so Claude Code and Codex can point at it directly. releasing weights publicly decides who can download a model, not who can afford to run one. frontier open models keep shipping, and running them still assumes a rented cluster. meanwhile there are over a hundred million consumer machines with discrete GPUs sitting mostly idle. closing that gap was never a hardware problem, and work like this is what turns open weights into something you can actually use. paper: repo: almost every idea in this post, from why memory bandwidth decides the outcome to why moving weights costs more than computing on them, comes straight out of how a GPU is built. I wrote a detailed primer on that. the article is quoted below.

336,634 views • 10 days ago

Web scraping will never be the same. (100% open-source visual search at scale) PixelRAG is a retrieval system that skips HTML parsing completely. Instead of scraping a page into text and embedding chunks, it screenshots the page and retrieves the image. A vision-language model reads the answer straight off the pixels. Why that matters: parsing is where web RAG quietly loses information. - A single HTML-to-text parser can drop 40%+ of a page. - Tables, charts, and layout get flattened or thrown out. - Swapping parsers alone can move accuracy ~10 points on the same docs. PixelRAG indexes the page a person actually sees. The team built a visual index of all of Wikipedia, 30M+ screenshots, and it still beats the strongest text RAG baseline by 18.1% on text-only QA. The repo also ships a Claude Code plugin that gives Claude eyes. It lets Claude screenshot any URL and read the rendered page instead of scraping the DOM. So you can hand it a live page, an arXiv paper, or your local site and ask what it actually looks like. One setup script. No MCP server, no backend. How the pipeline works: - Renders each document (web, PDF, image) to image tiles. - Embeds them with Qwen3-VL-Embedding, LoRA fine-tuned on screenshots. - Builds a FAISS index and serves a search API. A stronger reader model lifts accuracy with no re-indexing, since the index is just pixels. Everything is open-source under Apache-2.0. GitHub repo: Talking about RAG, I recently wrote an article on a new approach that makes retrieval much more efficient by cutting corpus size by 40x, reducing tokens per query by 3x, and improving vector search relevance by 2.3x. The article is quoted below.

Web scraping will never be the same. (100% open-source visual search at scale) PixelRAG is a retrieval system that skips HTML parsing completely. Instead of scraping a page into text and embedding chunks, it screenshots the page and retrieves the image. A vision-language model reads the answer straight off the pixels. Why that matters: parsing is where web RAG quietly loses information. - A single HTML-to-text parser can drop 40%+ of a page. - Tables, charts, and layout get flattened or thrown out. - Swapping parsers alone can move accuracy ~10 points on the same docs. PixelRAG indexes the page a person actually sees. The team built a visual index of all of Wikipedia, 30M+ screenshots, and it still beats the strongest text RAG baseline by 18.1% on text-only QA. The repo also ships a Claude Code plugin that gives Claude eyes. It lets Claude screenshot any URL and read the rendered page instead of scraping the DOM. So you can hand it a live page, an arXiv paper, or your local site and ask what it actually looks like. One setup script. No MCP server, no backend. How the pipeline works: - Renders each document (web, PDF, image) to image tiles. - Embeds them with Qwen3-VL-Embedding, LoRA fine-tuned on screenshots. - Builds a FAISS index and serves a search API. A stronger reader model lifts accuracy with no re-indexing, since the index is just pixels. Everything is open-source under Apache-2.0. GitHub repo: Talking about RAG, I recently wrote an article on a new approach that makes retrieval much more efficient by cutting corpus size by 40x, reducing tokens per query by 3x, and improving vector search relevance by 2.3x. The article is quoted below.

946,395 views • 2 months ago

Another blow to Anthropic! They spent months building what's now fully open-source. Anthropic recently put Claude inside Slack, where you can tag it in a channel. It reads the thread, breaks the task into steps, and posts the result back. The problem is that it only runs Claude and only in the channels Anthropic supports. Running your own agent there is harder. The reasoning, tool calls, and state management are mostly handled by the framework. Connecting that agent to a messaging platform is not. Moreover, each platform has a different integration: - Slack renders messages with Block Kit - Teams uses Adaptive Cards - and each has its own SDK, auth flow, and delivery model. If an agent needs to run on three platforms, one must write three separate integrations against the same agent logic. That overhead explains why most custom agents never get deployed to Slack, and why the ones that do are usually a single vendor's hosted assistant. The alternative is to keep the agent in one place and add a per-platform adapter that translates its output into each platform's native format. The agent is written once, and each channel requires just another output target instead of a separate build. CopilotKit open-sourced this full implementation in the Channels SDK. Essentially, any agent that implements AG-UI can run in a messaging platform in a few lines of code, like Slack, Teams, Discord, WhatsApp, and many more. Because the agent runs inside the thread, it has that conversation's context, so it can summarize the discussion, open a ticket, or route to the right person. It works with any backend, so LangGraph, CrewAI, Mastra, Google ADK, or a plain HTTP agent can connect through an existing endpoint. The same message can render as a Block Kit in Slack and as Adaptive Cards in Teams. In practice, the model and orchestration stay the same; it requires no migration or rewrite. It also handles human-in-the-loop approvals, persistence, and transcripts that carry state across platforms, so a thread started in Teams can continue in Slack. CopilotKit is open-source, and AG-UI is supported across every major agent framework, including LangGraph, CrewAI, Mastra, and Google ADK. Here's the repo: (don't forget to star it ⭐) The agent running in Slack no longer has to be a vendor's. It can be the one you already built. The video below shows this in action. Thanks to CopilotKit for working with me on this launch.

Another blow to Anthropic! They spent months building what's now fully open-source. Anthropic recently put Claude inside Slack, where you can tag it in a channel. It reads the thread, breaks the task into steps, and posts the result back. The problem is that it only runs Claude and only in the channels Anthropic supports. Running your own agent there is harder. The reasoning, tool calls, and state management are mostly handled by the framework. Connecting that agent to a messaging platform is not. Moreover, each platform has a different integration: - Slack renders messages with Block Kit - Teams uses Adaptive Cards - and each has its own SDK, auth flow, and delivery model. If an agent needs to run on three platforms, one must write three separate integrations against the same agent logic. That overhead explains why most custom agents never get deployed to Slack, and why the ones that do are usually a single vendor's hosted assistant. The alternative is to keep the agent in one place and add a per-platform adapter that translates its output into each platform's native format. The agent is written once, and each channel requires just another output target instead of a separate build. CopilotKit open-sourced this full implementation in the Channels SDK. Essentially, any agent that implements AG-UI can run in a messaging platform in a few lines of code, like Slack, Teams, Discord, WhatsApp, and many more. Because the agent runs inside the thread, it has that conversation's context, so it can summarize the discussion, open a ticket, or route to the right person. It works with any backend, so LangGraph, CrewAI, Mastra, Google ADK, or a plain HTTP agent can connect through an existing endpoint. The same message can render as a Block Kit in Slack and as Adaptive Cards in Teams. In practice, the model and orchestration stay the same; it requires no migration or rewrite. It also handles human-in-the-loop approvals, persistence, and transcripts that carry state across platforms, so a thread started in Teams can continue in Slack. CopilotKit is open-source, and AG-UI is supported across every major agent framework, including LangGraph, CrewAI, Mastra, and Google ADK. Here's the repo: (don't forget to star it ⭐) The agent running in Slack no longer has to be a vendor's. It can be the one you already built. The video below shows this in action. Thanks to CopilotKit for working with me on this launch.

243,461 views • 28 days ago

Sam Altman made the case for open-source harnesses in July. a month later, someone shipped it, and it's more efficient than most managed harnesses. here is the problem it was aimed at: a large share of your agent's token bill is the model rereading things it already read. that isn't the model's doing. the runtime around it decides what goes into every prompt and how often the model gets called. for example, an agent queries a CRM at step four and gets back 400 rows. those rows get piled up in the conversation history. by step nineteen, the model has to read those rows fifteen times unnecessarily, and every token read is billed at input rates. it happened because your harness assembled that prompt on every turn and kept the rows in it. that gives you two levers: how much context the harness carries forward, and how often it calls the model. there are four practical ways to keep the prompt from growing unnecessarily: → load tool schemas on demand. a server with 100 tools doesn't need to put all 100 into every prompt when the agent only calls two. → offload large results to disk. turn a large response into a short preview and a file path instead of replaying the entire result on every turn. → delegate to subagents. let a subagent spend thirty tool calls in its own context and return one summary to the root agent. → run toolchains in code. one script calls three tools, joins the results, and returns a table instead of three turns each dragging a full response. but reducing context is only half the job. you also need to control how often the model gets called. a good harness should avoid unnecessary planning, verification, and reflection when the work can be completed in fewer steps. TrueFoundry's open-source agent harness, TrueForge, is built around both of those controls. it sits between the model and the tools, deciding what goes into every prompt and when another model call is actually needed. it also breaks token usage down across the harness, skills, instructions, tools, and messages. DevRev's Enterprise-Bench is where this gets tested, on multi-step tasks of the kind where an agent pulls records from one system and reconciles them against another. TrueFoundry ran TrueForge there against Claude Managed Agents, both on the same model, and both finished the same number of tasks. the tie is the part that matters, because it means the gap underneath is not a quality tradeoff. TrueForge reached that score on close to a third of the tokens, with roughly 40% fewer trips back to the model. for the same result, that comes out around 2.7x cheaper than Claude Managed Agents. swapping in an open model made it sharper still. TrueForge with GLM-5.2 scored a little higher than either setup above, and the entire benchmark run cost about $3 at list prices. being open source matters beyond the license here. the model underneath can be swapped without rewriting the agent, and the whole thing can run inside your own environment when the data cannot leave it. all of this comes down to the runtime around the model, the context it carries, the tools it exposes, and how many times it goes back to the model. that is what a production harness actually owns. the full task list, the per-run numbers, and the MIT-licensed code are on GitHub: (don't forget to star 🌟) you can read more about the same in the article quoted below. thanks to the TrueForge team for working with me on this one.

Sam Altman made the case for open-source harnesses in July. a month later, someone shipped it, and it's more efficient than most managed harnesses. here is the problem it was aimed at: a large share of your agent's token bill is the model rereading things it already read. that isn't the model's doing. the runtime around it decides what goes into every prompt and how often the model gets called. for example, an agent queries a CRM at step four and gets back 400 rows. those rows get piled up in the conversation history. by step nineteen, the model has to read those rows fifteen times unnecessarily, and every token read is billed at input rates. it happened because your harness assembled that prompt on every turn and kept the rows in it. that gives you two levers: how much context the harness carries forward, and how often it calls the model. there are four practical ways to keep the prompt from growing unnecessarily: → load tool schemas on demand. a server with 100 tools doesn't need to put all 100 into every prompt when the agent only calls two. → offload large results to disk. turn a large response into a short preview and a file path instead of replaying the entire result on every turn. → delegate to subagents. let a subagent spend thirty tool calls in its own context and return one summary to the root agent. → run toolchains in code. one script calls three tools, joins the results, and returns a table instead of three turns each dragging a full response. but reducing context is only half the job. you also need to control how often the model gets called. a good harness should avoid unnecessary planning, verification, and reflection when the work can be completed in fewer steps. TrueFoundry's open-source agent harness, TrueForge, is built around both of those controls. it sits between the model and the tools, deciding what goes into every prompt and when another model call is actually needed. it also breaks token usage down across the harness, skills, instructions, tools, and messages. DevRev's Enterprise-Bench is where this gets tested, on multi-step tasks of the kind where an agent pulls records from one system and reconciles them against another. TrueFoundry ran TrueForge there against Claude Managed Agents, both on the same model, and both finished the same number of tasks. the tie is the part that matters, because it means the gap underneath is not a quality tradeoff. TrueForge reached that score on close to a third of the tokens, with roughly 40% fewer trips back to the model. for the same result, that comes out around 2.7x cheaper than Claude Managed Agents. swapping in an open model made it sharper still. TrueForge with GLM-5.2 scored a little higher than either setup above, and the entire benchmark run cost about $3 at list prices. being open source matters beyond the license here. the model underneath can be swapped without rewriting the agent, and the whole thing can run inside your own environment when the data cannot leave it. all of this comes down to the runtime around the model, the context it carries, the tools it exposes, and how many times it goes back to the model. that is what a production harness actually owns. the full task list, the per-run numbers, and the MIT-licensed code are on GitHub: (don't forget to star 🌟) you can read more about the same in the article quoted below. thanks to the TrueForge team for working with me on this one.

75,962 views • 13 days ago

Turn any PDF, image, DOCX, and PPTX into clean markdown. Parsing one PDF is easy. Parsing millions of them is where it gets tricky. Datalab just shipped Marker v2, a parsing pipeline that runs up to 23.7 pages/s on a single B200 GPU. Supports 90+ languages. 100% open-source.

Turn any PDF, image, DOCX, and PPTX into clean markdown. Parsing one PDF is easy. Parsing millions of them is where it gets tricky. Datalab just shipped Marker v2, a parsing pipeline that runs up to 23.7 pages/s on a single B200 GPU. Supports 90+ languages. 100% open-source.

54,457 views • 9 days ago

Anthropic won't like this open-source repo. It is going to cost LLM providers a lot of money. Every CI run of an AI app today sends real requests to providers like OpenAI or Anthropic. Like any other LLM call, this too gets billed at actual API rates. So for teams with high commit volumes, this accumulates into a meaningful chunk of API spend. One common hack devs use is that instead of invoking the LLM API, the test calls a fake local server that speaks the same API and returns a dummy response. The catch is that the dummy response is a copy of what the provider returned on the day it was saved, and providers keep adding fields and changing types. So the tests keep passing against a schema that's no longer valid, while the real integration breaks in production. A smart approach is now actually implemented in CopilotKit🪁's recently open-sourced aimock project. Every day, the repo's own CI sends a handful of requests to the real API and the same requests to the fake server, then compares both against the official client library's type definitions. Those are the only real API calls in the whole setup, and they run on the repo's own keys, not in anyone else's CI. A single team can push hundreds of commits a day, and thousands of teams are already doing that with coding agents. All of those runs stay offline, because one repo checks against the real API on everyone's behalf. When a check fails, a coding agent updates aimock's built-in response schema, the full test suite has to pass, and a patch version ships to npm. By simply upgrading the package, the corrected schema gets reflected in every project using it. The capability is not just limited to a single provider. The same server works for Claude, OpenAI, Gemini, Bedrock, Azure, Ollama, plus MCP tools, A2A agents, AG-UI event streams, vector DBs like Pinecone and Qdrant, and search, speech, image, and video endpoints. Here's the repo: (don't forget to star it ⭐) That said, mocking your API calls is one thing. AI engineers should also know how to test agents properly in the first place, which several teams still skip. I wrote a full walkthrough on that, covering build, testing, evals, tracing, and deployment. Read it below.

Anthropic won't like this open-source repo. It is going to cost LLM providers a lot of money. Every CI run of an AI app today sends real requests to providers like OpenAI or Anthropic. Like any other LLM call, this too gets billed at actual API rates. So for teams with high commit volumes, this accumulates into a meaningful chunk of API spend. One common hack devs use is that instead of invoking the LLM API, the test calls a fake local server that speaks the same API and returns a dummy response. The catch is that the dummy response is a copy of what the provider returned on the day it was saved, and providers keep adding fields and changing types. So the tests keep passing against a schema that's no longer valid, while the real integration breaks in production. A smart approach is now actually implemented in CopilotKit🪁's recently open-sourced aimock project. Every day, the repo's own CI sends a handful of requests to the real API and the same requests to the fake server, then compares both against the official client library's type definitions. Those are the only real API calls in the whole setup, and they run on the repo's own keys, not in anyone else's CI. A single team can push hundreds of commits a day, and thousands of teams are already doing that with coding agents. All of those runs stay offline, because one repo checks against the real API on everyone's behalf. When a check fails, a coding agent updates aimock's built-in response schema, the full test suite has to pass, and a patch version ships to npm. By simply upgrading the package, the corrected schema gets reflected in every project using it. The capability is not just limited to a single provider. The same server works for Claude, OpenAI, Gemini, Bedrock, Azure, Ollama, plus MCP tools, A2A agents, AG-UI event streams, vector DBs like Pinecone and Qdrant, and search, speech, image, and video endpoints. Here's the repo: (don't forget to star it ⭐) That said, mocking your API calls is one thing. AI engineers should also know how to test agents properly in the first place, which several teams still skip. I wrote a full walkthrough on that, covering build, testing, evals, tracing, and deployment. Read it below.

62,821 views • 15 days ago

Claude Code is now scary good at full-stack! I asked it to build a real-time weather intelligence dashboard with an interactive 3D globe, a forecasting layer that predicts weather 3 days ahead, and an anomaly detector that flags cities whose weather is behaving abnormally. It came back with a spinning globe that has a day/night cycle using NASA satellite imagery, city lights on the dark side, weather icons that switch between sun and moon based on local time, and a time travel slider that scrubs through 10 days of data. And when a city's weather breaks from its own normal, it pulses red (abnormally hot) or blue (abnormally cold) right on the globe, updating live and reflecting the anomaly state at any point you drag the slider to. Claude Code built the whole thing in a single session, including the backend, database, data pipeline, and frontend. For the database, I needed something fast for time-series workloads since the app ingests hourly weather readings across many cities and serves time-range queries on every slider interaction. I used Tiger Cloud by Tiger Data - Creators of TimescaleDB, which gives you managed TimescaleDB on the Postgres you already know. Claude Code connected to it through the Tiger CLI MCP server and set up the entire backend directly: - Provisioned the database service - Created hypertables for time-partitioned weather storage - Set up continuous aggregates for pre-computed rollups - Built the data ingestion pipeline and the full NextJS + ThreeJS frontend The time travel slider queries thousands of rows on every position change. On a regular Postgres table, this would require manual partitioning and index tuning to stay fast as data grows. TimescaleDB partitions the data by timestamp automatically, so each query only hits the relevant time chunk. Continuous aggregates serve the trend charts, the forecast layer, and the anomaly baselines from pre-computed rollups instead of rescanning raw data on every request. The video below shows the final build in action, and I worked with the Tiger Data team to put this together. Tiger CLI is open-source (Apache 2.0) and works with Claude Code, Cursor, Codex, Gemini CLI, and VS Code. To try this yourself: → Sign up for Tiger Cloud (I have shared the link in the replies). It gives you $1,000 free credits (no card needed) → Install Tiger CLI: curl -fsSL https(:)//cli(.)tigerdata(.)com | sh → Run tiger mcp install claude-code → Give Claude Code a prompt and let it build sign-up here: My co-founder also wrote a detailed article on this. The article is quoted below.

Claude Code is now scary good at full-stack! I asked it to build a real-time weather intelligence dashboard with an interactive 3D globe, a forecasting layer that predicts weather 3 days ahead, and an anomaly detector that flags cities whose weather is behaving abnormally. It came back with a spinning globe that has a day/night cycle using NASA satellite imagery, city lights on the dark side, weather icons that switch between sun and moon based on local time, and a time travel slider that scrubs through 10 days of data. And when a city's weather breaks from its own normal, it pulses red (abnormally hot) or blue (abnormally cold) right on the globe, updating live and reflecting the anomaly state at any point you drag the slider to. Claude Code built the whole thing in a single session, including the backend, database, data pipeline, and frontend. For the database, I needed something fast for time-series workloads since the app ingests hourly weather readings across many cities and serves time-range queries on every slider interaction. I used Tiger Cloud by Tiger Data - Creators of TimescaleDB, which gives you managed TimescaleDB on the Postgres you already know. Claude Code connected to it through the Tiger CLI MCP server and set up the entire backend directly: - Provisioned the database service - Created hypertables for time-partitioned weather storage - Set up continuous aggregates for pre-computed rollups - Built the data ingestion pipeline and the full NextJS + ThreeJS frontend The time travel slider queries thousands of rows on every position change. On a regular Postgres table, this would require manual partitioning and index tuning to stay fast as data grows. TimescaleDB partitions the data by timestamp automatically, so each query only hits the relevant time chunk. Continuous aggregates serve the trend charts, the forecast layer, and the anomaly baselines from pre-computed rollups instead of rescanning raw data on every request. The video below shows the final build in action, and I worked with the Tiger Data team to put this together. Tiger CLI is open-source (Apache 2.0) and works with Claude Code, Cursor, Codex, Gemini CLI, and VS Code. To try this yourself: → Sign up for Tiger Cloud (I have shared the link in the replies). It gives you $1,000 free credits (no card needed) → Install Tiger CLI: curl -fsSL https(:)//cli(.)tigerdata(.)com | sh → Run tiger mcp install claude-code → Give Claude Code a prompt and let it build sign-up here: My co-founder also wrote a detailed article on this. The article is quoted below.

37,269 views • 11 days ago

How to build a 1-person AI company that: - Runs locally - 100% open-source - No human employees, all agents - Real-time collaboration via email Multi-agent orchestration is not new. Plenty of frameworks already let agents hand off tasks, run in parallel, and talk to each other. So the interesting question is not whether agents can collaborate. It is what structure you use to make them collaborate. The common approach is to wire a graph of nodes and edges and reason about the plumbing yourself. It works, but you are learning a new abstraction just to describe who does what. There is a coordination structure we have trusted for a hundred years already: an organization. Every company runs the same way. People have roles, roles have reporting lines, and work moves up and down that chart without anyone relaying each message by hand. Map that onto agents and the whole thing gets intuitive. You lay out an org chart, each agent fills one role, you talk to the person at the top, and the org sorts out the work between them. You already know how a company works, so you already know how to run one here. There is no new abstraction to learn. That is exactly what Alook does. Each agent is a live Claude Code or OpenCode session with a defined role, a reporting line, and its own email inbox. The agents coordinate over email, the same way a team would. And it all runs locally through a runtime on your own machine, so nothing leaves your setup. You bring your own agent too. Claude Code and Codex both work, and if you would rather stay fully open source and local, OpenCode works the same way. To show how this feels in practice, I set up three agents as a small sales team. Vi is the one I talk to. I hand Vi a goal, and Vi routes the work down the chart. Neile runs prospect research. Vi passes the target criteria, and Neile reports back a ranked list of names, roles, and companies, each with a suggested angle and a confidence score. Lliane runs outreach. Vi hands over the messaging angle and follow-up cadence, and Lliane reports back on emails sent, responses received, and any deal that needs escalation. I never relay a message between them. Neile and Lliane report to Vi, and Vi updates me in one place. The whole thing is open source and self-hosted, so it runs on your machine with your own agents. Give the repo a star if you want to follow where it goes: I also wrote a full walkthrough on building your own AI company with it, from a blank org chart to a running job. The article is quoted below. Cheers! :)

How to build a 1-person AI company that: - Runs locally - 100% open-source - No human employees, all agents - Real-time collaboration via email Multi-agent orchestration is not new. Plenty of frameworks already let agents hand off tasks, run in parallel, and talk to each other. So the interesting question is not whether agents can collaborate. It is what structure you use to make them collaborate. The common approach is to wire a graph of nodes and edges and reason about the plumbing yourself. It works, but you are learning a new abstraction just to describe who does what. There is a coordination structure we have trusted for a hundred years already: an organization. Every company runs the same way. People have roles, roles have reporting lines, and work moves up and down that chart without anyone relaying each message by hand. Map that onto agents and the whole thing gets intuitive. You lay out an org chart, each agent fills one role, you talk to the person at the top, and the org sorts out the work between them. You already know how a company works, so you already know how to run one here. There is no new abstraction to learn. That is exactly what Alook does. Each agent is a live Claude Code or OpenCode session with a defined role, a reporting line, and its own email inbox. The agents coordinate over email, the same way a team would. And it all runs locally through a runtime on your own machine, so nothing leaves your setup. You bring your own agent too. Claude Code and Codex both work, and if you would rather stay fully open source and local, OpenCode works the same way. To show how this feels in practice, I set up three agents as a small sales team. Vi is the one I talk to. I hand Vi a goal, and Vi routes the work down the chart. Neile runs prospect research. Vi passes the target criteria, and Neile reports back a ranked list of names, roles, and companies, each with a suggested angle and a confidence score. Lliane runs outreach. Vi hands over the messaging angle and follow-up cadence, and Lliane reports back on emails sent, responses received, and any deal that needs escalation. I never relay a message between them. Neile and Lliane report to Vi, and Vi updates me in one place. The whole thing is open source and self-hosted, so it runs on your machine with your own agents. Give the repo a star if you want to follow where it goes: I also wrote a full walkthrough on building your own AI company with it, from a blank org chart to a running job. The article is quoted below. Cheers! :)

169,957 views • 1 month ago

This is the DeepSeek moment for Voice AI. Chatterbox Turbo is an MIT-licensed voice model that beats ElevenLabs Turbo & Cartesia Sonic 3! - <150ms time-to-first-sound - Voice cloning from just 5-second audio - Paralinguistic tags for real human expression 100% open-source.

This is the DeepSeek moment for Voice AI. Chatterbox Turbo is an MIT-licensed voice model that beats ElevenLabs Turbo & Cartesia Sonic 3! - <150ms time-to-first-sound - Voice cloning from just 5-second audio - Paralinguistic tags for real human expression 100% open-source.

468,703 views • 8 months ago

Karpathy said something you'll regret ignoring: "You are still responsible for your software, just as before. You are not allowed to introduce vulnerabilities because of vibe coding. " He said it while drawing the line between vibe coding and agentic engineering. Agents write more of the code now, but none of that takes the responsibility off you. The assumption underneath that is that a careful enough reader catches the problem. But some failures don't show up in anything there is to read. For instance, a common fear with a RAG agent is that it could hallucinate when a question asks something outside its corpus. But such cases are actually well handled by any competent model now. If nothing in the retrieved context looks relevant, there's no material to build an answer on. Instead, the majority of failures originate when the retrieved context has partial coverage. The retrieval pipeline returns context that's topically correct but doesn't cover the full question, and the model completes the remainder from parametric knowledge. There are no token-level labels in the output to tell what was generated using retrieved context and what came from weights. Both are streamed the same way. Detecting this for production-grade apps needs a metric written for it, one that's also aligned with principles of agentic engineering. And the solution is actually implemented in the eval skill that comes with Google’s Agents CLI. I described the concern to Claude Code in plain English. It read the agent's code, came back with a plan I approved. It then reported that no built-in metric isolates the behaviour and wrote a custom rubric called corpus_abstention. It assigned a single categorical verdict per case rather than aggregating everything into one score, since the built-in raters regenerate their rubrics each run and leave no stable number to trend. → GROUNDED_ANSWER → CORRECT_ABSTENTION → UNGROUNDED_ANSWER (answered entirely from outside knowledge) → MIXED_LEAKAGE (grounded, but slips in one unsupported claim) → WRONG_ABSTENTION (refused something the docs actually covered) After this, it automatically generated 33 scenarios partitioned by where the failure could occur, like: - in-corpus - off-domain - out-of-corpus but plausibly answerable - boundary cases where the topic is covered, but a specific detail isn't. The baseline score was 19 of 33. - Off-domain passed 3 of 3, as expected. - But 6 of 15 in-corpus cases retrieved the right document, cited it correctly, answered accurately, and added a claim the source never made. The root cause was one line in the agent's instruction: "If you already know the answer to a simple question and no document lookup is needed, you may respond directly without citations." The eval skill helped flag this, and then Claude removed it and forced retrieval on every question. This took the suite to 30 of 33, and ungrounded answers went from 6 to 0. The full recording of my run is below, and I worked with the Google Cloud team on this. Agents CLI GitHub repo → (don't forget to star 🌟) I wrote up the full build covering all six steps from install to enterprise registration. It includes the eval scorecard, the instruction loophole the eval caught before deployment, and what the deployment process actually looks like end-to-end. Read it below.

Karpathy said something you'll regret ignoring: "You are still responsible for your software, just as before. You are not allowed to introduce vulnerabilities because of vibe coding. " He said it while drawing the line between vibe coding and agentic engineering. Agents write more of the code now, but none of that takes the responsibility off you. The assumption underneath that is that a careful enough reader catches the problem. But some failures don't show up in anything there is to read. For instance, a common fear with a RAG agent is that it could hallucinate when a question asks something outside its corpus. But such cases are actually well handled by any competent model now. If nothing in the retrieved context looks relevant, there's no material to build an answer on. Instead, the majority of failures originate when the retrieved context has partial coverage. The retrieval pipeline returns context that's topically correct but doesn't cover the full question, and the model completes the remainder from parametric knowledge. There are no token-level labels in the output to tell what was generated using retrieved context and what came from weights. Both are streamed the same way. Detecting this for production-grade apps needs a metric written for it, one that's also aligned with principles of agentic engineering. And the solution is actually implemented in the eval skill that comes with Google’s Agents CLI. I described the concern to Claude Code in plain English. It read the agent's code, came back with a plan I approved. It then reported that no built-in metric isolates the behaviour and wrote a custom rubric called corpus_abstention. It assigned a single categorical verdict per case rather than aggregating everything into one score, since the built-in raters regenerate their rubrics each run and leave no stable number to trend. → GROUNDED_ANSWER → CORRECT_ABSTENTION → UNGROUNDED_ANSWER (answered entirely from outside knowledge) → MIXED_LEAKAGE (grounded, but slips in one unsupported claim) → WRONG_ABSTENTION (refused something the docs actually covered) After this, it automatically generated 33 scenarios partitioned by where the failure could occur, like: - in-corpus - off-domain - out-of-corpus but plausibly answerable - boundary cases where the topic is covered, but a specific detail isn't. The baseline score was 19 of 33. - Off-domain passed 3 of 3, as expected. - But 6 of 15 in-corpus cases retrieved the right document, cited it correctly, answered accurately, and added a claim the source never made. The root cause was one line in the agent's instruction: "If you already know the answer to a simple question and no document lookup is needed, you may respond directly without citations." The eval skill helped flag this, and then Claude removed it and forced retrieval on every question. This took the suite to 30 of 33, and ungrounded answers went from 6 to 0. The full recording of my run is below, and I worked with the Google Cloud team on this. Agents CLI GitHub repo → (don't forget to star 🌟) I wrote up the full build covering all six steps from install to enterprise registration. It includes the eval scorecard, the instruction loophole the eval caught before deployment, and what the deployment process actually looks like end-to-end. Read it below.

82,546 views • 1 month ago

i just built a 4-agent software team. everything runs from Telegram and gets managed on a kanban board. a project manager who plans the work, a backend developer, a frontend developer, and a tester. the PM reads a goal, breaks it into linked tasks, and assigns each to the right agent. the thing that makes them a team instead of four strangers is a shared kanban board. every task is a row that survives crashes, and when an agent finishes, it writes a summary of what it built and what the next agent needs to know. the next agent reads that summary before it starts. so the frontend developer never has to guess the API shape, and the tester knows exactly what to verify. the hardest part was not the coordination. it was building an agent that could actually act like a backend engineer. a backend engineer stands up a database, wires auth, manages storage, deploys functions, and keeps all of it consistent while the rest of the team builds on top. an agent doing this from scratch drowns. it burns its context window remembering which tables exist and which endpoint it created three steps ago, and the work degrades fast. so the backend agent needs a backend built for agents, not for humans clicking through a dashboard. that is where InsForge came in. it is an open-source, agent-native backend, and i added it to my backend developer agent as a skill. a skill is a step-by-step guide that teaches the agent how to do a specific kind of work. with InsForge installed, the agent stopped improvising infrastructure and followed a reliable path: create the project, define the database, set up auth, deploy functions. to test the whole team, i had them build a working Google Docs clone, AI features included. the backend agent spun up the full service on its own. database tables, user auth, document handling, and edge functions running real TypeScript, all in one dashboard. the frontend agent read that summary and built the UI on top of it, and the tester closed the loop. the result was a backend an agent could reason about end to end, instead of one it kept getting lost inside. if you are building an AI backend engineer, InsForge is worth a look, it's 100% open-source. InsForge GitHub: (don't forget to star 🌟) the full article on Hermes Kanban: Mission Control for your Agents is quoted below.

i just built a 4-agent software team. everything runs from Telegram and gets managed on a kanban board. a project manager who plans the work, a backend developer, a frontend developer, and a tester. the PM reads a goal, breaks it into linked tasks, and assigns each to the right agent. the thing that makes them a team instead of four strangers is a shared kanban board. every task is a row that survives crashes, and when an agent finishes, it writes a summary of what it built and what the next agent needs to know. the next agent reads that summary before it starts. so the frontend developer never has to guess the API shape, and the tester knows exactly what to verify. the hardest part was not the coordination. it was building an agent that could actually act like a backend engineer. a backend engineer stands up a database, wires auth, manages storage, deploys functions, and keeps all of it consistent while the rest of the team builds on top. an agent doing this from scratch drowns. it burns its context window remembering which tables exist and which endpoint it created three steps ago, and the work degrades fast. so the backend agent needs a backend built for agents, not for humans clicking through a dashboard. that is where InsForge came in. it is an open-source, agent-native backend, and i added it to my backend developer agent as a skill. a skill is a step-by-step guide that teaches the agent how to do a specific kind of work. with InsForge installed, the agent stopped improvising infrastructure and followed a reliable path: create the project, define the database, set up auth, deploy functions. to test the whole team, i had them build a working Google Docs clone, AI features included. the backend agent spun up the full service on its own. database tables, user auth, document handling, and edge functions running real TypeScript, all in one dashboard. the frontend agent read that summary and built the UI on top of it, and the tester closed the loop. the result was a backend an agent could reason about end to end, instead of one it kept getting lost inside. if you are building an AI backend engineer, InsForge is worth a look, it's 100% open-source. InsForge GitHub: (don't forget to star 🌟) the full article on Hermes Kanban: Mission Control for your Agents is quoted below.

122,548 views • 2 months ago

I just built my own multi-agent GTM research assistant! (it finds the reason to reach out before it writes a single message) Cold outreach usually fails on timing, not on wording. By the time you find out that a target company raised a round or hired a new data leader, the window has closed and your message reads like every other cold email in the inbox. So the research has to happen before the writing, and it has to run across the whole target list at once. Today, we're building a system where you drop in target company names and three agents handle the rest. Here's how it works: ↳ Agent 1 searches recent news and pulls trigger events for each company ↳ Agent 2 finds people at those companies and enriches their career background ↳ Agent 3 joins the two and writes a ready-to-send message per contact ↳ Results come back ranked by how strong the trigger is ↳ The whole pipeline runs inside a Streamlit UI The order matters more than the agent count. The writer agent runs last and receives the trigger event and the contact's background as its input, so it never starts from a blank company name. If the news agent finds nothing recent for a company, there is nothing to write from, and that company drops down the ranking instead of producing a generic message. Tech stack: ↳ Seltz as the data layer, running the news scope and the people scope over the same target list ↳ CrewAI to orchestrate the three agents in sequence ↳ Streamlit to host the interface Here's why I picked this stack: Seltz maintains its own web index instead of wrapping a search engine, so the news scope and the people scope return structured results you can join on the company name. That join is the whole system. A trigger event with no contact attached is not actionable, and a contact with no trigger gives you nothing to say. Get started here: CrewAI keeps the handoffs explicit. Each agent receives the previous agent's output as context instead of starting its own search from scratch, so the message writer already knows both what happened at the company and who it is writing to. Streamlit keeps the target list, the pipeline run, and the final ranked output in one place, which makes the intermediate agent output easy to inspect when a message comes out wrong. Find all the code and everything you need to run this app in the studio: I also wrote a comprehensive article that covers this entire idea and how to replicate this for your own use case in more detail. The article is quoted below.

I just built my own multi-agent GTM research assistant! (it finds the reason to reach out before it writes a single message) Cold outreach usually fails on timing, not on wording. By the time you find out that a target company raised a round or hired a new data leader, the window has closed and your message reads like every other cold email in the inbox. So the research has to happen before the writing, and it has to run across the whole target list at once. Today, we're building a system where you drop in target company names and three agents handle the rest. Here's how it works: ↳ Agent 1 searches recent news and pulls trigger events for each company ↳ Agent 2 finds people at those companies and enriches their career background ↳ Agent 3 joins the two and writes a ready-to-send message per contact ↳ Results come back ranked by how strong the trigger is ↳ The whole pipeline runs inside a Streamlit UI The order matters more than the agent count. The writer agent runs last and receives the trigger event and the contact's background as its input, so it never starts from a blank company name. If the news agent finds nothing recent for a company, there is nothing to write from, and that company drops down the ranking instead of producing a generic message. Tech stack: ↳ Seltz as the data layer, running the news scope and the people scope over the same target list ↳ CrewAI to orchestrate the three agents in sequence ↳ Streamlit to host the interface Here's why I picked this stack: Seltz maintains its own web index instead of wrapping a search engine, so the news scope and the people scope return structured results you can join on the company name. That join is the whole system. A trigger event with no contact attached is not actionable, and a contact with no trigger gives you nothing to say. Get started here: CrewAI keeps the handoffs explicit. Each agent receives the previous agent's output as context instead of starting its own search from scratch, so the message writer already knows both what happened at the company and who it is writing to. Streamlit keeps the target list, the pipeline run, and the final ranked output in one place, which makes the intermediate agent output easy to inspect when a message comes out wrong. Find all the code and everything you need to run this app in the studio: I also wrote a comprehensive article that covers this entire idea and how to replicate this for your own use case in more detail. The article is quoted below.

13,305 views • 7 days ago

The Hermes Desktop App is insanely good. It's now the best way to run AI agents on your computer. Here's the full setup, start to finish. Enjoy! Chapters: 00:00 - why Hermes desktop is a game changer 00:46 - downloading and installing the desktop app 01:47 - picking your model and provider 02:52 - sessions, settings, and gateway connection 04:08 - adding a custom MCP server 04:58 - memory and context (the three-tier system) 06:52 - connecting to Telegram 09:45 - skills and tools 11:06 - the skills hub (built-in and community skills) 11:50 - creating a custom skill 14:18 - artifacts 15:11 - going from one to many agents (profiles and personas) 18:12 - agents working as a team (Hermes Kanban) 19:01 - outro I have also written an article on Hermes masterclass, the same is quoted below.

The Hermes Desktop App is insanely good. It's now the best way to run AI agents on your computer. Here's the full setup, start to finish. Enjoy! Chapters: 00:00 - why Hermes desktop is a game changer 00:46 - downloading and installing the desktop app 01:47 - picking your model and provider 02:52 - sessions, settings, and gateway connection 04:08 - adding a custom MCP server 04:58 - memory and context (the three-tier system) 06:52 - connecting to Telegram 09:45 - skills and tools 11:06 - the skills hub (built-in and community skills) 11:50 - creating a custom skill 14:18 - artifacts 15:11 - going from one to many agents (profiles and personas) 18:12 - agents working as a team (Hermes Kanban) 19:01 - outro I have also written an article on Hermes masterclass, the same is quoted below.

103,813 views • 2 months ago

Everyone is sleeping on this new OCR model! - 85.9% (sota) on olmocr bench - 90+ language support w/benchmarks - 4B model (down from 9B) - Full layout information - Extracts + captions images and diagrams - Strong handwriting, math, form, table support 100% open-source.

Everyone is sleeping on this new OCR model! - 85.9% (sota) on olmocr bench - 90+ language support w/benchmarks - 4B model (down from 9B) - Full layout information - Extracts + captions images and diagrams - Strong handwriting, math, form, table support 100% open-source.

168,552 views • 5 months ago

Software engineers are going to love this! I found an open-source error monitoring agent that scans production logs, finds the root cause, and sends a Slack message with full context before you even notice something broke. Cuts down production downtime by 95%! Check this:

Software engineers are going to love this! I found an open-source error monitoring agent that scans production logs, finds the root cause, and sends a Slack message with full context before you even notice something broke. Cuts down production downtime by 95%! Check this:

181,414 views • 6 months ago

I just built my own OCR app powered by Google's Gemma 3. It runs 100% locally on my computer using Ollama and extracts text from images as structured markdown. The code is open-source, and you can find it in the next tweet.

I just built my own OCR app powered by Google's Gemma 3. It runs 100% locally on my computer using Ollama and extracts text from images as structured markdown. The code is open-source, and you can find it in the next tweet.

477,685 views • 1 year ago

Claude Skills might be the biggest upgrade to AI agents so far! Some say it's even bigger than MCP. I've been testing skills for the past 3-4 days, and they're solving a problem most people don't talk about: agents just keep forgetting everything. In this video, I'll share everything I've learned so far. It covers: > The core idea (skills as SOPs for agents) > Anatomy of a skill > Skills vs. MCP vs. Projects vs. Subagents > Building your own skill > Hands-on example Skills are the early signs of continual learning, and they can change how we work with agents forever! Here's everything you need to know:

Claude Skills might be the biggest upgrade to AI agents so far! Some say it's even bigger than MCP. I've been testing skills for the past 3-4 days, and they're solving a problem most people don't talk about: agents just keep forgetting everything. In this video, I'll share everything I've learned so far. It covers: > The core idea (skills as SOPs for agents) > Anatomy of a skill > Skills vs. MCP vs. Projects vs. Subagents > Building your own skill > Hands-on example Skills are the early signs of continual learning, and they can change how we work with agents forever! Here's everything you need to know:

286,274 views • 10 months ago

Make Claude Code 10x more powerful. Claude-Mem is a free plugin to persist memory across Claude sessions. It captures tool usage, so you always start where you left off. Endless Mode allows 95% token reduction & 20x more tool use before context exhaustion. 100% open-source.

Make Claude Code 10x more powerful. Claude-Mem is a free plugin to persist memory across Claude sessions. It captures tool usage, so you always start where you left off. Endless Mode allows 95% token reduction & 20x more tool use before context exhaustion. 100% open-source.

184,476 views • 8 months ago

Everyone is sleeping on this new OCR model! dots-ocr is a new 1.7B vision-language model that achieves SOTA performance on multilingual document parsing. - Supports 100+ languages - Works with both images and PDFs - Handles text, tables, formulas seamlessly 100% open-source.

Everyone is sleeping on this new OCR model! dots-ocr is a new 1.7B vision-language model that achieves SOTA performance on multilingual document parsing. - Supports 100+ languages - Works with both images and PDFs - Handles text, tables, formulas seamlessly 100% open-source.

252,205 views • 1 year ago

A 100% open-source alternative to n8n! Sim is a drag-and-drop UI for creating powerful AI agent workflows: - Runs locally on your machine - Works with local LLMs I built a stock market research agent & connected it to Telegram in minutes. Here's a step-by-step guide:

A 100% open-source alternative to n8n! Sim is a drag-and-drop UI for creating powerful AI agent workflows: - Runs locally on your machine - Works with local LLMs I built a stock market research agent & connected it to Telegram in minutes. Here's a step-by-step guide:

176,601 views • 8 months ago