Baidu just open-sourced an OCR model that reads entire... 40-page documents in one shot. It's called Unlimited-OCR. 3 billion parameters but only 500 million active during inference. Runs 100% locally on your machine. Why this matters: traditional OCR tools chop documents page by page. Tables that span two pages break. Reading order gets lost. Cross-page context disappears. Unlimited-OCR processes the whole document at once. 32K context window. Text, formulas, tables, reading order all preserved across pages. Output comes out as clean structured Markdown. → 93% accuracy on the standard benchmark. +6 points over the baseline. → Error rate stays below 0.11 even past 40 pages. → Multilingual out of the box. → 2.12 million downloads on Hugging Face last month. 14,600 GitHub stars. For context: Amazon Textract, Google Cloud Vision, and Azure Document Intelligence all charge per page. This runs locally for free.show more

A peanut-sized Chinese model just dethroned Gemini at reading... documents. GLM-OCR is a 0.9B parameter vision-language model. It scores 94.62 on OmniDocBench V1.5, ranking #1 overall. For context, it outperforms models 100x its size. 100% open-source. It works in two stages. 1. A layout engine detects every region in a document. 2. Each region gets read in parallel. The model predicts multiple tokens per step instead of one. That's what makes it so fast at small size. It handles things most OCR tools struggle with: > Complex tables and nested layouts > Handwritten text and stamps > Math formulas and code blocks > Mixed image-and-text documents You can run it locally through Ollama. It fits on edge devices with limited compute. Every expensive OCR API just got a free competitor.show more

A peanut-sized Chinese model just dethroned Gemini at reading... documents. GLM-OCR is a 0.9B parameter vision-language model. It scores 94.62 on OmniDocBench V1.5, ranking #1 overall. For context, it outperforms models 100x its size. 100% open-source. It works in two stages. 1. A layout engine detects every region in a document. 2. Each region gets read in parallel. The model predicts multiple tokens per step instead of one. That's what makes it so fast at small size. It handles things most OCR tools struggle with: > Complex tables and nested layouts > Handwritten text and stamps > Math formulas and code blocks > Mixed image-and-text documents You can run it locally through Ollama. It fits on edge devices with limited compute. Every expensive OCR API just got a free competitor.show more

1.7 billion free tokens per month. A month ago... i showed you how to route claude code through free providers. someone just shipped the cleanest version of this setup yet… it's called Freellmapi 13,400+ stars on github, MIT licensed, takes 2 minutes to install. what it does: stacks the free tiers of 16 different LLM providers behind one local API. point claude code, codex, or cursor at that one endpoint, and it automatically routes your calls across all 16 free pools. The 16 providers it covers: Google, Groq, Cerebras, Mistral, OpenRouter, GitHub Models, Cloudflare, Cohere, NVIDIA, HuggingFace, Ollama Cloud, Kilo, Pollinations, LLM7, OVH, and OpenCode Zen. if you sign up to all 16 and add your free API keys, you get roughly 1.7 billion free tokens per month combined. ▫️ How to install (one command) curl -fsSL bash this runs the whole thing locally on your machine through Docker. once it's up, open paste your provider keys on the Keys page, and grab the unified API key from the dashboard. that's the key you point your apps at. With this, claude code stops hitting your monthly cap because every prompt routes through the 16 free pools instead of your paid plan. and if one provider rate-limits mid-conversation, freellmapi falls over to the next one automatically so your session never breaks. repo: Free, MIT-licensed, runs on your laptop or a $5 VPS.show more

you're paying $20/mo for something your $500 GPU can... already do. Gemma 4 26B A4B QAT MoE + Hermes Agent running on a single RTX 4060 (8GB VRAM). Built a vision capable, 100% free, 100% local, private AI assistant that lives in my Chrome browser. No API keys. No cloud. No subscriptions. 100% vibe coded. 0% handholding. It has full context of whatever's on my screen can answer questions, summarize pages, extract data, and see images. Same local model handles everything, no external calls, ever. keep reading for the model and hermes agent tips i learnt while building this locally. Here's the exact setup for anyone running local LLMs on 6-8 GB VRAM: llama.cpp server flags (on my NVIDIA RTX 4060 8gb VRAM): -m gemma-4-26B-A4B-it-qat-UD-Q4_K_XL.gguf --cache-type-k q8_0 --cache-type-v q8_0 -c 150000 --port 8080 Throughput with quantization: Prefill: 200-250 tokens/sec Decode: 20-25 tokens/sec reduce context if oom on 6 gb vram card. Key learnings: - Quantize KV cache to q8 for faster prefill/decode. Prefill goes from 100-150 (unquantized) to 200-250 tok/s (q8). - But watch out, once actual context grows past ~50k tokens on high entropy workloads, q8 KV quantization can cause hallucinations. Low entropy workloads are mostly unaffected. If you see it happening, drop the quantization. This is common across all local models. - In Hermes Agent settings -> Memory & Context, bump compression threshold from default 0.5 to 0.7. Default triggers way too frequent context compression and eats time. Up next: add persistent memory, web search, tool calling, streaming output and whatever you suggest. Running a 26B MoE with vision + 150k context window on 8GB VRAM would've sounded impossible 6 months ago. Works the same on the NVIDIA RTX 3060 Ti, 3070, 4060 Ti, 5060, 2080, or any 8GB card. VRAM is the only requirement. Local AI agents are closer than people think. You just need to know where the knobs are. Model's Unsloth quant hugging face link in the comments. Have you tried Hermes agent by Nous Research yet? What are you building with local LLMs? Drop it below, let's see what this community is shipping.show more

This Chinese developer launched Llama 70B locally on a... MacBook on a plane and for a full 11 hours without internet ran client projects. He was sitting by the window on a transatlantic flight with a MacBook Pro M4 with 64 GB of memory. WiFi on board cost $25 for the flight. He declined. No cloud API, no connection to Anthropic or OpenAI servers, no internet at all. Just a local Llama 3.3 70B on bf16 and his own orchestrator script. The model runs through llama.cpp. Generation speed, 71 tokens per second. Context around 60,000 tokens. Memory usage, 48.6 GiB out of 64. Battery at takeoff, 3 hours 21 minutes. And he gave the orchestrator this system prompt before takeoff: "You are an offline orchestrator running on a single MacBook. There is no network. The only resources you have are local files in /Users/dev/work, the Llama 70B inference server at localhost:8080, and a battery budget of 3 hours 21 minutes. Process the queue at /Users/dev/work/queue.jsonl (one client task per line). For each task: draft → run local evals → save artefact to /Users/dev/work/done/. Save context checkpoints every 12 tasks so you can resume after a battery swap. Stop only on empty queue or when battery drops below 5%." So the system knows exactly what resources it is running on. It knows it has no connection to the outside world for the next 11 hours. It knows it has finite memory and a finite battery. It knows the human will not intervene until the plane lands. The system runs in 1 loop. Takes a task from the queue, runs it through inference, saves the artifact, writes a checkpoint. Task after task, just like that. And only when the battery drops below 5% does the orchestrator automatically pause, waits for the laptop to switch to the backup power bank, and continues from the last checkpoint. Here is what the system actually writes in his log during the flight: "saved context checkpoint 8 of 12 (pos_min = 488, pos_max = 50118, size = 62.813 MiB)" "restored context checkpoint (pos_min = 488, pos_max = 50118)" "prompt processing progress: n_tokens = 50 / 60 818" "task 37016 done | tps = 71 s tokens text → /Users/dev/work/done/proposal_westside.md" Outside the window, clouds, blue sky, and no WiFi. On the tray, 1 MacBook, an open terminal on 2 screens, and an inference server on localhost. From what I have observed, this is the cleanest offline AI workflow I have seen in the past year: 11 hours of flight, $0 for WiFi, and the entire client queue closed before landing.show more

I just ran Gemma 4 31B on @CerebrasSystems at... 1,800+ tokens/sec and it's multimodal. For context: that's 35x faster than a typical GPU endpoint, and the first token (reasoning included) lands in 1.5 seconds. This isn't a benchmark slide, I recorded the inference live. Prompt I used: "Create a simulation of an iPhone. Include at least one working dummy note taking app, a functional notification pulldown, high quality graphics, single HTML file, any libs via CDN." - Generation time: 3 seconds. - Notes app worked. - Notification panel worked. - Rendered first try. This is what wafer-scale inference unlocks, not just "faster," but a different category of product. When generation is this fast, you stop waiting and start iterating in real time. Why this matters: Gemma 4 31B is Google DeepMind's flagship open weight model, Apache 2.0 licensed, dense (not MoE), and built for efficiency over raw parameter count. It scores close to Claude Haiku 4.5 on the Artificial Analysis Intelligence Index (30 vs 29) but runs ~18x faster on Cerebras. It's also the first multimodal model on Cerebras's platform, meaning you can now feed it screenshots, documents, charts, and UI states at wafer scale speed. # Applications I'm most excited about: - Screenshot → Insight: Drop in a dashboard or document screenshot, get structured findings back instantly. no waiting, no batching. - Live UI generation: Full interactive interfaces (like my iPhone sim) generated and rendered in under 2 seconds. - Screenshot -> Patch: Feed it a broken UI + console error, get a minimal code fix and verification steps back. - Computer use & agentic loops: See -> reason -> act - verify, fast enough to keep a human in the loop instead of waiting on the model. - Long context summarization: Full research reports condensed into decision ready summaries you can read and requery in one sitting. The bigger unlock isn't the speed number itself, it's that agentic and multimodal loops (see -> reason -> output -> tool call -> verify -> retry) finally run in real time instead of feeling sluggish. As Logan Kilpatrick (Logan Kilpatrick) put it: "If every model was doing 2,000 tokens per second, you wouldn't build the same product and just have it be faster, you'd build different products." Gemma 4 31B is live now on Cerebras Inference Cloud in public preview. If you're building multimodal, agentic, or real time apps, this is worth testing today. What would you build with such insane inference throughput?show more

I am the Under Secretary for Strategic Resource Integration.... We sit inside the Executive Office of the President. You won't find us on an org chart. We requested it that way. You think an office like mine spins up in a crisis. We've been open since January. Venezuela wasn't a war. Venezuela was the beta. Maduro proved the model the morning we took delivery of him. The President called it "working out brilliantly for both." He's right. It works brilliantly for both of us, and the customer will come around to the "both" on their own time. Kharg Island is the rollout. A third the size of Manhattan, and 90 percent of Iran's crude runs through it. 2 million barrels a day, one valve. I have prayed my whole career for a country that legible. 53 billion in revenue. 11 percent of their GDP. All of it funneling through a rock you could walk across before lunch. I kept the Venezuela playbook. I changed the classification banner and let find-and-replace do Caracas to Tehran. That was the plan. That was the whole plan. Page 14 still says Caracas. No one on the distribution list has ever reached page 14. They initial page 1. Page 1 is projected revenue. It goes up and to the right, the only direction this town can read. The Energy Secretary testified he was "unaware." Under oath, to Congress, the man who runs American energy. Good. He was never read into the compartment. The fewer names on the account, the cleaner the account. He runs the department. He does not run this. We don't say invasion. We say onboarding. Every onboarding has friction. The Marines off the coast are not an occupying force. They're field engagement. Iran said an invasion of its islands would "shatter all restraint." I logged it as customer feedback and filed it. The file is full. Engaged customers submit the most feedback. It's the quiet ones who churn. And Iran can't churn. That's the part I'm proudest of. We built the account with no exit. No competitor to switch to. We are the market, and the market is now total. Total is my word. I workshopped it. "Seize" scares people. "Total control of their markets" tests at 71, because nobody pictures a kid from Ohio on a beach when you say markets. We moved 22 tankers through Hormuz at night with the lights off. The President said he wanted to announce it so badly it hurt, and he held it a month anyway. That is the most discipline this administration has ever shown. 100 million barrels, he said. The fact-checkers say the math doesn't hold. They're junior. The number was never the deliverable. The number is the soft launch. He said he loves the inflation. People clutched their receipts. 90 a barrel instead of 250 is the discount, and 250 is a price I invented so this one would feel like mercy. How long will we be there? The President answered that. "We had to be there for a while." A while is my favorite contract length. It renews on its own. No court can find the end date. He wondered out loud whether America has the stomach for it. I don't need the country's stomach. I need its signature on page 1. He called the people who object stupid. I'd defend them. They're not stupid. They just read to page 14. Last thing. Annex C. The out-year targets. 3 more, redacted, marked pre-decisional. I can't tell you which 3. Total control of their markets. Nobody in that room asked which market. The crude still sails east, to China, the same as yesterday. We don't own the oil. We own the valve, the fear, and the word "total." That was always the only product.show more

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.show more

A 16-year-old in Austin made $49,200 in six months... while every law firm in his city was busy counting Google reviews that nobody under 30 reads anymore. He walked into a law firm and asked the paralegal to search for the practice on Perplexity. The paralegal laughed and pointed at 400 five-star reviews on Google. He said, "Just do it." Perplexity had never heard of them. Here is what the kid understood that the paralegal did not. Google reviews are a ranking signal inside Google's algorithm. Perplexity runs its own crawler. It does not care how many stars you have on a platform it is not reading. It pulls from legal directories, bar association profiles, Yelp, structured schema data, and third-party citations. A firm can sit at the top of the Google Local Pack with 847 reviews and have zero citation presence inside the AI systems that 500 million users query every month. As of February 2026, the overlap between pages ranking in Google's top 10 and pages cited inside AI-generated answers had collapsed from 76 percent to under 20 percent. Two entirely different systems. Almost nobody in legal had noticed. The paralegal thought the reviews were the proof. The kid saw they were the blind spot. So he built a $1,200 audit. The deliverable is a single document. He opens Perplexity, ChatGPT, and Claude. He types the firm's practice area and city. He screenshots what comes back. Then he runs the same search on every competitor in the market. He maps which firms are being named, where the citations are coming from, and what data signals are missing from the ones that do not appear. The finding is almost always identical: No Foursquare listing No attorney schema or LegalService markup beyond the default WordPress install Bar association profile unlinked from the main site Attorney bios with no verifiable credentials structured for machine reading NAP inconsistent across the seven directories that Perplexity actually indexes A firm charging $450 an hour that ChatGPT cannot confidently recommend because it cannot verify the address matches across three platforms. Less than 5 percent of local businesses have done this work as of 2026. In legal, the number is closer to zero. He charges $1,200 to show them exactly where they do not exist. Then he quotes them the fix. He walked out of the first firm with a check. That firm referred him to two others before the week was over. Those two referred three more. He has never made a cold call. He has never run an ad. He does not have a website. 41 firms in six months. $49,200 in revenue. He is 16. From what I have observed, the arbitrage here is not technical. It is perceptual. Law firms spent a decade optimizing for a system that is no longer the first place their clients look. The 16-year-old simply walked in and showed them the new one.show more

I went a little overboard with Codex last week... and burned through my entire weekly allowance in two days. Luckily, my quota reset today. Otherwise, I’m not sure what I would’ve done. It got me thinking: instead of asking one large model to handle everything from start to finish, why not let a stronger model plan the project and review the work, while a model built for execution handles the day-to-day implementation? So I tried it. The result was better than I expected. I used GPT-5.6 Sol in Codex as the decision-maker, then ran Ling-3.0-flash from Ant Ling inside OpenCode as the execution engine. Together, they built a small 3D farming game. Before writing any code, I had Codex create four documents: SPEC.md defined the product scope and the lines we couldn’t cross. ARCHITECTURE.md laid out the isometric coordinate system, state machine, and module boundaries. TASKS.md broke the project into small jobs Ling could tackle one at a time. ACCEPTANCE.md explained how each step would be tested and what “done” actually meant. Then I gave Ling a very straightforward role: You are the execution model for this project. Read all four documents before you begin. Work only on the task assigned for this round. When you’re done, run typecheck, test, and build. If anything fails, read the error, fix it, and run the checks again. Do not move on to the next task early. Ling handled dependency installation, project structure, strict TypeScript configuration, test setup, and a production build in 6 minutes and 3 seconds. It ran into issues with the Vite test config, a TS6310 error, and a missing jsdom dependency along the way. Instead of stopping at the first error, it kept reading the logs and fixing the problems until all three checks passed. The speed was honestly hard to believe. If you exclude the time spent waiting on tools, it was producing more than 100 tokens per second. That made the whole development loop feel noticeably faster. After this experiment, I’m planning to keep using the same workflow. If the task is small, there’s no reason to call an expensive planning model for every single step. If the task is large, handing the entire project to a Flash model in one prompt isn’t a great idea either. The setup that makes more sense to me is: Use a more capable model such as Codex to explore the project, make architectural decisions, and break the work down. Put the constraints into specs, schemas, types, and tests instead of leaving them buried in chat history. Give Ling-3.0-flash a steady stream of clear, verifiable implementation tasks. Report bugs with structured context and actual error logs, rather than saying, “It still doesn’t work.” Bring Codex back in for architecture reviews, visual checks, and changes that affect multiple parts of the project. The point of this setup isn’t to give AI a big “build the whole project” button. It’s to turn software development into a pipeline with a much more sensible cost structure: Codex figures out the plan, sets the boundaries, and catches problems. Ling-3.0-flash moves quickly, calls tools reliably, and works through well-defined tasks at scale. For agent workflows that involve lots of repetitive edits, production tasks, and tool calls, this may be a more practical answer than simply using the biggest model for everything.show more

I told you to claim your free 16GB NVIDIA... GPU for learning Local LLMs. Now I’m going to show you how to double its inference speed without touching the hardware. Google Colab gives you an enterprise grade NVIDIA Tesla T4 GPU for free, roughly 4 hours every single day. It is the absolute perfect sandbox for learning AI engineering, testing inference flags, and pushing massive context windows. The local AI timeline is moving way too fast. If you aren't using Multi Token Prediction (MTP) yet, you are leaving massive performance on the table. I just pushed DeepMind’s Gemma 4 26B to 64.9 t/s on this exact free tier. Let's look at the raw benchmark data running on an Ubuntu Linux environment with the latest compiled llama.cpp binaries and quantized GGUFs from Unsloth via HuggingFace: # Qwen 3.5 9B (Dense): Base: [ Prompt: 626.7 t/s | Generation: 21.0 t/s ] With MTP: [ Prompt: 539.1 t/s | Generation: 24.8 t/s ] # Gemma 4 26B QAT (MoE): Base: [ Prompt: 634.2 t/s | Generation: 48.3 t/s ] With MTP: [ Prompt: 572.1 t/s | Generation: 64.9 t/s ] If you are paying attention, this single Colab notebook reveals 3 massive observations about the current state of local LLMs: # 1. The MTP Speedup (Software Overclocking) Standard autoregressive decoding guesses one token at a time. MTP acts like a highly optimized, built in speculative decoder. It predicts multiple future tokens at once and the main model verifies them in parallel. The result? Zero accuracy loss and a massive throughput increase. Gemma jumped from 48 to 65 t/s just by flipping a flag. # 2. The MoE Paradox (Bigger is Faster) How does a 26B parameter model absolutely destroy a 9B model in raw speed on the exact same hardware? Architecture. Qwen 3.5 9B is a dense model. it activates all 9 billion parameters for every single token. Gemma 4 26B is a Mixture of Experts (MoE) model. It routes data efficiently, activating only 4B parameters per token. You get the reasoning capabilities of a 26B model with the compute cost of a 4B model. 3. Thinking Efficiency When I ran the exact same complex prompt on both models, the larger MoE spent significantly fewer "thinking" tokens to arrive at the correct answer. A smarter model doesn't just give better answers; it gets to the point faster, saving you compute cycles and preserving your context window. # Want to run this yourself? Here are the exact llama.cpp CLI commands. For Qwen (MTP is baked into the main model): ./llama-cli -m Qwen3.5-9B-UD-Q4_K_XL.gguf -p "Explain quantum computing." -n 2000 -c 8000 -ngl 99 -fa on --spec-type draft-mtp --spec-draft-n-max 4 --spec-draft-p-min 0.7 For Gemma (Using a separate lightweight draft model): ./llama-cli -m gemma-4-26B-A4B-it-qat-UD-Q4_K_XL.gguf --model-draft mtp-gemma-4-26B-A4B-it.gguf -p "Explain quantum computing." -n 2000 -c 8000 -ngl 99 -fa on --spec-type draft-mtp --spec-draft-n-max 4 --spec-draft-p-min 0.7 Stop waiting for a $3,000 rig. Boot up Colab, pull these models, and start building your stack. I’ve put together a completely free, cell by cell Google Colab notebook that automates this entire workflow so you can test it yourself in 5 minutes and learn. Link to the notebook is in the comments below. Experiemt with different MTP parameters, context windows and post your results in the comments.show more

HERMES AGENT HAS A SECOND BRAIN. 1,100+ KNOWLEDGE FILES.... AUTO-LINKED. SELF-IMPROVING. GROWING EVERY NIGHT. THIS IS THE OBSIDIAN GRAPH BEHIND IT. every dot = one knowledge file (markdown) every line = one wiki-link between files every color = one category (skills, notes, decisions, sources, entities) HOW IT BUILDS ITSELF: Hermes ships with a bundled LLM Wiki skill. based on Andrej Karpathy's pattern. unlike RAG (rediscovers knowledge from scratch every query), the wiki compiles knowledge once and keeps it current. when you feed the agent a source: → it reads the content → writes a structured markdown page → auto-links to every related existing page → flags contradictions with previous entries → updates all affected pages one source in. multiple connections created. the graph grows denser with every entry. WHAT FEEDS THE WIKI: → articles and URLs you find interesting → meeting transcripts → PDF documents and research papers → conversation history from Hermes sessions → Claude Code and Codex session history → Slack logs, email threads, saved notes → YouTube transcripts → raw text dropped into a _raw/ folder the obsidian-wiki package supports multi-agent ingest from Hermes, Claude Code, Codex, OpenClaw, Pi, Windsurf, and ChatGPT exports. install: pip install obsidian-wiki obsidian-wiki setup --vault ~/wiki AUTOMATE THE GROWTH: set cron jobs to feed the wiki overnight: "every day at 9am, check for new meetings. ingest transcripts into the wiki." "every week, check arXiv for new papers in [niche]. summarize and file into the wiki." "every day, ingest today's Hermes sessions into the wiki under session-history." month 1: 50 entries. scattered. month 3: 300+ entries. cross-referenced. month 6: 1,000+ entries. the agent surfaces patterns you never searched for. WHY OBSIDIAN: the wiki is plain markdown files. no database. no lock-in. open it in Obsidian for graph view: → nodes show knowledge density → links show how ideas connect → clusters reveal your strongest domains → orphan nodes reveal gaps Hermes writes from a VPS. Obsidian reads on your laptop. obsidian-headless syncs without a GUI. agent writes from the server, you browse on your device. FOUR MEMORY LAYERS: Layer 1: memory.md + user.md (~2,200 + 1,375 chars. short-term.) Layer 2: SQLite with FTS5 (full session transcripts. searchable.) Layer 3: external providers (Mem0, SuperMemory, Honcho. optional.) Layer 4: Obsidian wiki via LLM Wiki skill (unlimited. compounding. the long-term brain.) layers 1-3 handle memory. layer 4 handles knowledge. the graph in this post is layer 4. SETUP: set in Desktop app, Dashboard, or config.yaml: WIKI_PATH=~/wiki OBSIDIAN_VAULT_PATH=~/wiki first run: Hermes asks for your domain. answer with your niche. the skill builds SCHEMA.md with tag taxonomy. after that: "index this into my wiki: [URL or text]" the wiki grows. the graph densifies. the agent gets smarter because the knowledge base got smarter. full 15 levels breakdown in the article 👇show more

my 8 GB VRAM gaming laptop is absolutely going... to hate me for this. but I still did it. ran a 31b dense model (Gemma 4 31b Q4) with only 8 GB VRAM last week I ran Gemma 4 26B A4B a mixture of experts model on my RTX 4060 and hit 25–28 tokens/sec using llama.cpp's new MTP support. smooth. snappy. but MoE has a secret: it only activates 4B parameters per token despite having 26B total. that's why it flies. so the real question started haunting me. what if I throw a full, no tricks, every parameter fires on every token, 31B DENSE model at the same machine? # Hardware: GPU: NVIDIA RTX 4060, 8 GB VRAM RAM: 16 GB CPU: Intel Core i7 H Laptop. Gaming. Modest. The model: gemma-4-31B-it-qat-UD-Q4_K_XL.gguf (model's unsloth huggingface link in the comments) This is Google DeepMind's flagship dense model in the Gemma 4 family that can run on single consumer GPU. It packs a hybrid attention architecture, supports up to 256K context natively, and is QAT (Quantization Aware Training) optimized, meaning it retains far more quality than standard post training quants at the same bit depth. This is NOT the MoE. This is 31 BILLION dense parameters, every single one of them loaded. # the flags I used: -m gemma-4-31B-it-qat-UD-Q4_K_XL.gguf -cnv --spec-type draft-mtp --spec-draft-model mtp-gemma-4-31B-it.gguf --spec-draft-n-max 8 --spec-draft-p-min 0.6 -c 6000 -v Multi Token Prediction (MTP) is still active here. Separate draft GGUF required, same as the 26B setup. # Results: → Decode: ~3 tokens/sec → Prefill: ~2 tokens/sec → Context: 6000 tokens → Hardware crying quietly in the corner: yes so is 3 tps actually usable? For real time back and forth chat? Not ideal. You're not having a fluid conversation at 3 tps. but slow ≠ useless. And this is where it gets genuinely interesting. think about how senior devs actually work in a real team. But when something is architectural, deeply complex, or needs serious reasoning? they walk down the hall and escalate to the senior. That's exactly the local AI agent architecture this unlocks: → Fast orchestrator model (Gemma 4 26B MoE at 25+ tps) handles routing, simple queries, tool calls, memory. The junior dev. → Gemma 4 31B dense is the senior, called only when the fast model genuinely hits a wall. Hard multi step reasoning. Complex code generation. Deep architectural decisions. The agentic loop stays fast. Only the hard hops touch the 31B. That's a legitimate production grade local AI architecture on a budget hardware. (requires 2 8gb gpus) other workflows where 3 tps is completely fine: - overnight batch jobs. summarize documents, extract structured data, review code. Fire it off. Sleep. wake up to results. - One shot deep reasoning - Silent code audit loops, you write and test, the 31B reviews diffs and flags issues in the background between your sprints - Any workflow where output quality > output speed A few weeks ago, nobody was running a 30B+ dense model on a single consumer GPU with 8 GB VRAM. At all. Now we're doing it on an Intel i7-H gaming laptop with a NVIDIA RTX 4060, thanks to llama.cpp + QAT quants + MTP speculative drafting. Google DeepMind said the Gemma 4 31B targets "consumer GPUs and workstations." They were not exaggerating. The hardware bar to run serious frontier class models locally keeps dropping. the tools are here. the models are here. you just have to be willing to abuse your laptop a little. what workflows would you actually run on a local 3 tps 31B dense model? genuinely curious. drop it below.show more

Dawah is deception. Why are you leaving out the... context SeerahPro ? You know right after this he says he doesn’t remember well what he wrote in that entry on Muhaymin. Why leave that part out? He even goes on to say that section seems to be non-committal and he needs to go back and re-read it. So since you don’t want to show every one the full context, I’ll do it, and expose how you are lying now and lied in the debate. Sinai doesn’t help your claim that the Quran believes the previous scriptures were textually corrupted. He says he doesn’t think the Quran believes the previous scriptures were corrupted in the deep past and that they are unrecoverable (Clip 1). Now let’s look at the short clip you pulled in context (Clip 2). First, what he says does not affirm why you cited him to begin with, which is that the Quran is supposed to be a supreme authority and determine which parts of our scriptures are corrupt and which are not. In the very clip you pulled he says, “The Quran is claiming to be able to sort of determine the correct reading of these previous scriptures and to remedy misreading and misunderstandings.” So by arbiter he means it is not determining what parts of our scriptures are true or corrupted, it is only an arbiter in determining meaning. You just buried yourself by citing this, because you implied Sinai was saying the Quran is an arbiter for determining textual corruption, which he doesn’t say. It’s like you didn’t even pay attention to what you pulled. That is embarrassing. Second, right after this (in what you left out) the interviewer reminds him he entertained two possibilities in his book and was noncommittal. Sinai says he doesn’t remember what he said and was probably leaving open possibilities, which is what I already had to correct you on. If you go back to the book, once again, he said on page 707 essentially what I said, “Muhaymin, meanwhile, is derived from Syriac mhaymnā (or conceivably from its equivalent in some other form of Aramaic), a passive participle of the verb haymen and meaning ‘trustworthy, faithful, loyal.’” After this all he does in the book is mention other possibilities and the one you want is more contingent on non-canonical readings. But even if that interpretation is correct (which Sinai confirms in this interview is his view) it doesn’t help you, because you misunderstood what Sinai means by ‘arbiter.’ In Clip 3 he first says, “So I need to reread that entry. I guess it seems to be very non-committal.” But remember you said (Clip 4), “In his 'Key Terms of the Quran' he explicitly says that Muhaymin means the Quran is the arbiter of the contents.” That is incorrect. He did not explicitly say that. Why did you lie? In Clip 3 he goes on to say “But I think entrusted with authority can mean, authorized to sort of authoritatively interpret and determine especially theological significance.” So you don’t even understand what Sinai means by “arbiter.” He doesn’t mean what you mean, that the Quran is determining which parts of our scriptures are true and corrupted, but that it is an arbiter of meaning. The fact that you still do not see this means you are beyond delusional. You never should have cited Sinai because he doesn’t help you at all. The funny thing is we were already planning to go live tomorrow to review this interview because he doesn’t help you at all and unintentionally helps with the Islamic Dilemma.show more

This guy cracked the code on AI girlfriend monetization... using real-time technology and now pulls $76,000 per month from one Instagram profile without ever showing his real face or hiring an actual model. He got tired of watching creators split 80 percent of revenue with agencies while their competitors ran 24/7 chat operations with zero burnout, so he built a system that generates hyperrealistic AI influencer content using motion capture and synthetic face generation instead of photographers, makeup artists, or Miami beach rentals. His monthly profit hit $76,455 last month from just 90.4K followers and organic short-form traffic, while traditional creators cap out at $15K after paying 40 percent platform fees and $2,000 monthly for content production teams. Here is the exact breakdown: → Real-time face swap technology becomes the only tool you need, but most people butcher the setup by skipping gesture synchronization in the first 10 seconds → Character design comes first, and if you mess this up nothing saves it. Stick to approachable features (freckles, natural makeup, warm smile) because that is where parasocial engagement lives → Profile building is not random. You craft one consistent AI persona that repeats across all content so your audience recognizes the girl → You are picking who your subscriber projects onto, not who looks unattainable. That is your retention baked into the face → Motion capture runs before generation, and this is what kills the uncanny valley effect that destroys engagement in 3 seconds → You mirror your own gestures through webcam: confused shrug, hand raise, lean-in shock, peace sign wave. The AI mask tracks every micro-movement and applies it to the generated face in real time → Batching is the move 91 percent skip: same room setup, multiple emotion sequences, one recording session. → The system generates 7 to 10 TikToks before dinner, while traditional creators test 3 per week and wonder why their conversion rates are stuck at 0.4 percent The economics are stupid: each video costs him $0 in talent fees, pulls 2 million views organically, converts at 2 percent into 1,800 clicks to private platforms at $10 to $15 subscription with $40 to $60 backend PPV per fan. That is $76,455 profit per month, while real creators pay $5,000 for production and net $22,000 after platform cuts. The key move nobody talks about: you cannot skip the natural gesture library. If you generate the AI face without mirroring your own spontaneous reactions first, the avatar moves like a CGI render. The eye contact breaks. The smile timing lags. The whole thing screams and your retention dies at 2.1 seconds. His system records him doing the exact confusion-to-delight emotional arc first, so the AI mask inherits human timing, natural eyebrow raises, and spontaneous energy that reads as a real girl reacting to comments, not a scripted advertisement. One Instagram profile generated 12 variants of the same "how I afford this lifestyle" hook in 40 minutes with different outfits, different lighting setups, different trending audios, and found the winner in 96 hours without spending $8,000 on influencer collaborations. They were previously paying $1,200 per UGC creator and burning $6,400 per week on content that plateaued at 60K views. Now they spend $0 for 12 variants and their cost per subscriber dropped from $48 to $11. Agencies now panic because their entire margin was built on model exclusivity, and this removes the human dependency. The outfit changes between clips like a wardrobe swap filter. The lighting matches bedroom authenticity. The hand gestures sync with emotional beats. No casting call. No model contract. No location scouting. Just a webcamera, a real-time face swap AI, and the discipline to batch-test emotional hooks before you commit traffic spend to one persona.show more

I am the person at Hut 8 who designed... the American Bitcoin partnership. The structure is elegant. We gave the Trump family 20% of a publicly traded mining company. They contributed zero capital. Zero infrastructure. Zero employees. Zero operational experience. Zero risk exposure. They contributed a name. Per our partnership agreement, that is consideration. Twenty percent of our equity for access to the most valuable retail distribution channel in American finance. "It has to have 'America,'" Eric said in our first meeting. "And it has to have 'Bitcoin.'" He said this twice. Both times he pointed at the whiteboard. There was nothing else on the whiteboard. I realized then that he understood the product better than I did. The product is not bitcoin. The product is the belief. The entire business model. Two words and a surname. I wrote the term sheet on one page. The lawyers billed for forty. We call that alignment of incentives. Forty pages means they believed in the durability of the arrangement. We mine bitcoin at an all-in cost of approximately $90,000 per coin. Hash rate, power purchase agreements, ASIC depreciation, facility lease, headcount, Coinbase Prime interest — $90,000. Bitcoin trades at $77,000. Every coin we mine loses $13,000. Negative unit economics on every block reward. Eric tells investors we mine at $57,000. He strips out depreciation, SG&A, and the debt service. I asked him once if he understood what depreciation meant. He said it means when things go down. I said yes. He said: "But the stock goes up." I said yes. His only contractual obligation. Salesmanship. Per the partnership agreement, salesmanship is Eric's sole KPI. Technically, he is a fiduciary to shareholders. On paper, his vesting is tied to total comp benchmarks. We run the rigs. He runs the ticker. Asset-light. The company at peak reached a $13.2 billion valuation. Two employees. That is the entire headcount. One is our CEO Mike Ho, who is simultaneously Hut 8's Chief Strategy Officer. He reports to us at Hut 8 on Monday mornings and reports to American Bitcoin shareholders on Tuesday mornings. Dual-reporting structure. Very efficient. The other employee manages Eric's media calendar. $6.6 billion per headcount. We call this capital efficiency. 70% of our bitcoin did not come from mining. It came from selling stock. Retail investors purchase American Bitcoin shares at 50 times book value because the name contains "America" and "Bitcoin" and "Trump" is in the filing and they believe, with the quiet religious certainty of people who have never read a balance sheet in their lives, that a company named American Bitcoin is underwritten by something more substantial than two words and a surname. We take their cash and buy bitcoin on Coinbase at spot. Lodge it on the balance sheet. Call ourselves a mining company. We do mine. At a loss. Technically, the earnings are negative per our Q4 filing. The margin lives in the distance between what the stock costs them and what the bitcoin costs us. The stock is down 92% from peak. Investors have lost approximately $500 million. One of them posted on the shareholder subreddit that he moved his daughter's 529 into American Bitcoin at $14. It trades under $2. He said he believed in the mission. That means he believed in the name. The name performed exactly as designed. Eric's net worth went from $190 million to $280 million. Asset-light. We pledged 3,090 bitcoin as collateral against a Coinbase Prime custody loan. We have mined 1,800. The LTV ratio is inverted. If bitcoin compresses or the loan accelerates, every coin mined since inception could be forfeit by August 2027. All of it. Gone. Liquidation event. I explained this in a memo to Eric. Bullet points. Large font. He asked if the stock could go up before August. I said probably. He said that was fine. He said he'd handle it. Salesmanship. Eric told the press he launched American Bitcoin because banks were "debanking" the Trump family. I checked. JPMorgan refinanced $700 million in Trump Organization debt during the identical period. But debanking is better salesmanship than refinancing. The narrative inflates the stock price. The stock price generates the bitcoin. The bitcoin secures the loan. The loan generates cash. Every link in the chain is a product I built or a story Eric told. Asset-light. I orchestrated the celebrity endorsements. Tyler Winklevoss. Anthony Scaramucci. Grant Cardone. We call this pipeline development. Each broadcast the stock to their audiences during the run-up. The stock collapsed afterward. The celebrities did not lose money. Their audiences lost money. I never mentioned that we hemorrhage $13,000 per coin mined. I told them it was asset-light. They understood immediately. They are also asset-light. Eric cannot legally serve as a corporate officer in the state of New York. A judge barred him for two years. Civil fraud. So his title is not CEO. Not officer. Not executive. His contractual role is salesmanship. He cannot manage the company. He can sell it. One distinction. $90 million in personal net worth gained. Asset-light. Our CEO lives in the UAE. He held discussions with ADQ and TAQA, Abu Dhabi's sovereign wealth apparatus. The same sovereign apparatus that paid $500 million for 49% of World Liberty Financial, the family's other crypto operation. This is the same Abu Dhabi whose semiconductor imports the administration greenlit over national security objections. I did not design World Liberty Financial. I designed the mining subsidiary that feeds into it. Separate projects. Complementary revenue streams. Eric runs salesmanship for both. I admire the portfolio diversification. I gave Eric 20% of a company for free, a company with real miners and real facilities and real electricity bills that I built over seven years in Alberta and Texas and Ontario, and in exchange he gave me access to every American who hears "America" and "Bitcoin" in the same sentence and reaches for their brokerage app without checking whether the company mines at a profit or at a loss or at all. They drove the stock to a $13.2 billion market capitalization. We bought bitcoin with the proceeds. They lost $500 million. We kept the bitcoin. Eric kept $90 million. I kept the apparatus that manufactures both. Everybody got what they paid for. Asset-light means we carry nothing. Not the miners. Not the facilities. Not the risk. Not the losses. The investors carry those. We carry the bitcoin. Asset-light.show more

vPay offshore accounts and physical cards have been getting... field-tested IRL for a while now, and we’ll open them to the public as soon as we’re fully confident in the UX. But before offshore accounts go public, I want to address a few points: Some might point out that - vPay isn’t the first crypto card - vPay doesn’t have the lowest fees - So why choose vPay instead of the Coinbase 🛡️ Card or MetaMask 🦊 Card or KAST or or Tria, or any of the other big names? Now to address: Privacy | The biggest differentiator that sets vPay completely apart is Private Banking. The majority of the crypto card providers on the market use Rain infra. Even if you’ve never heard of them, that's what your favorite "NeoBank" uses. And due to their legal jurisdictions, they will report your finances to authorities since they're CRS and FACTA compliant. We are not. As an OmniBank, we work with different banking partners, and although KYC is required to use our services, our offshore banks are non-CRS and non-FACTA. Tax reporting is the responsibility and choice of the user. Offshore Accounts vs Physical Cards | I've tried to highlight this a few times so far. vPay has 3 offerings on the banking side of things. Virtual cards - live now. Physical cards - coming Q1 2026. The first two are similar to what everyone else on the market offers. The offshore accounts are not. which are coming this week. They allow unlimited spending, ATM withdrawals, and international SWIFT transfers, which very few “Neobanks” provide. Offshore accounts are coming this week. Self-Custody | We're not 100% non-custodial yet, as that is near impossible at the moment but it's something we're working towards. And we try to keep the users' self-custodial wallets in the loop as much as possible for maximum control. Those who have tried the vPay app know that almost every move asks for permission from their wallet, and we always encourage users to keep their funds in their non-custodial wallets until the very last moment, since our top-ups usually only take seconds to a minute to process. Fees | All of the card providers mentioned above either raised millions from VCs or in presales or have a huge org backing them. We have neither. vPay was self-funded and community-owned since day 1, launched under Virtuals Protocol Genesis V1 launch model, an objectively bad launch model and hugely unfavorable toward project teams. So even though vPay has been generating revenue and profitable from early on, we do not have the luxury of offering 0% fees yet, since they're mostly a marketing gimmick paid for by millions in VC money and not a sustainable business model for early-stage companies. What we're working towards instead, is true co-ownership of vPay and revenue-share with users. OmniBank vs NeoBank | I’m not a fan of the term “NeoBank.” It implies just a bank, but make it crypto. That’s not vPay. Our goals have always been clear: A) Anything and everything users need to do with their money and assets, both Web2 and Web3, all in one hub. Powered by a constellation of partner agents. The cards and the bank accounts are just the foundation. B) To eventually build independent financial rails for crypto and decouple from the chokehold of Visa/Mastercard. vLink is the first step toward this vision. This turned out to be a rather long tweet, but context matters. Questions and feedback welcome in replies or DMs. See you all with your vPay vCards very soon.show more

A Jane Street quant sat down across from me... at Dandelion I was grinding through a loss streak. Three red windows in a row. BTC 5-min bot on screen. She clocked it from the next table. "Is that Polymarket? Why is your oracle pinned" I told her. Chainlink. Resolves every 5 minutes. Binance and Coinbase are the real price. She moved her matcha over without asking. "You're trading oracle lag. We did this on ETH perps in 2022. Regulators shut us down in six weeks" Not perps. Binary markets. Nobody regulates a 5-minute window. 86 million Polymarket trades. Every fill. Every book snapshot. Every resolution. "You wrote a scorer on top of this" I didn't. Claude did. One prompt. 21 days of tick data. I asked what setup has the highest follow-through. Binance and Coinbase both cross the target by $50. Same direction. Chainlink hasn't printed yet. 94% of the time the oracle catches up inside 2 minutes. "So you're front-running a feed that can't front-run you back" Yeah. She pulled out a notebook. Actual paper. Wrote something. Circled it twice. "At the desk we called this a stale quote trade. Died the minute oracles went sub-second" I told her Chainlink on Polymarket still runs on a 5-minute update. That's the whole game. A green fill landed. +$52. "What's the second layer" Order book imbalance. First 10 levels. First 90 seconds. Above 1.8 buyers are loading. Below 0.55 sellers are breaking it. Retail doesn't show up until t+240. Three files. Entry scorer. Exit trigger. Settlement router. Claude rebuilds the scorer every Sunday from the week's logs. "You're letting it rewrite its own strategy" Exactly. "That's the part my old risk team would have lost sleep over" The exit is what keeps it alive. 0.75 shares. Never resolution. Polygon settles in 1 to 3 seconds and half the fills miss in the last minute. Early exit locks 70% of max. Redeploy next window. "55 out of 400 a day" How did you know. "Kelly fraction on a 71% hit rate with that payoff geometry. You'd be insane to trade more" She wasn't wrong. My setup: Claude API - $20/mo Hetzner VPS - $5/mo poly_data - free polymarket-trade-engine - free Polymarket/agents - free 30 days. 1,847 trades. 71% win rate. +$14,200. Copytrade here: Sharpe 2.84. Max DD -$640. Avg hold 3:12. She closed her notebook. "I make more than this in a bad afternoon at the desk. But the desk doesn't let me copy-paste a scorer from Claude on a Sunday" I told her that's the whole point. She stared at the screen for a minute. "Can I follow this wallet" Already live. The article was up the next morning. Her partner DMed me by lunch. Three lines. "Saw your post. My entire prop group is reading it. We'd like to talk" I told him the post is the talk. Everything's in it. Nothing left to gatekeep.show more

AgentLinter is here! Is your agent sharp & secure?... I built AgentLinter, a linter for and agent config files. Here's why. Whether you're vibe-coding or agent-coding, your AI's output quality comes down to one thing: how well you wrote your But managing these files properly? Way harder than it looks. 🎯 The Silent Failure Problem Vague instructions like "write good code" let the agent interpret however it wants. Output gets inconsistent, but nothing throws an error. The failure is silent. Anthropic's own docs say write "Use 2-space indentation" not "Format code properly." But as the file grows, spotting these with your eyes alone is nearly impossible. 🔐 The Security Problem People hard-code API keys and tokens directly into or and commit them, way more often than you'd think. AgentLinter stats show 1 in 5 workspaces has exposed credentials. .gitignore doesn't catch secrets buried inside markdown files. 💥 The Consistency Problem Multiple config files = contradictions. says "be a friendly assistant," says "concise, direct tone." The agent gets confused. references files that don't exist. Past 5 files, these conflicts triple. So I thought: is code. Code has ESLint. Why doesn't this have a linter? 🔍 What AgentLinter Does It diagnoses your agent config across 8 categories: 1) Structure: file organization 2) Clarity: instruction specificity 3) Completeness: missing definitions 4) Security: exposed secrets 5) Consistency: cross-file contradictions 6) Memory: session handoff 7) Runtime Config: gateway/auth settings 8) Skill Safety: dangerous shell commands & injection patterns Each scored 0–100 with concrete fix suggestions. Write "be helpful" and it tells you to specify response length, tone, and format. Find an API key? Instant CRITICAL alert to rotate. 🔒 Privacy-First & 100% Local Everything runs on your machine. Files never leave. Only the results are shared, and you can turn that off in settings. This matters — these files can contain system prompts, security rules, and personal context. Fully open source, MIT license, 100% free. 🛠️ Multi-Tool Support Works with Claude Code, Cursor, Windsurf, and Clawdbot. Detects for project mode, or clawdbot.json for agent mode and adjusts diagnostics automatically. 🚀 Get Started with one line npx agentlinter Node.js 18+, no config needed. Run it, check your score, fix what needs fixing. Happy vibe-coding & happy agent life! 🤙 Website: Github:show more

One player I've really grown to like over the... last several months is OF Kane Kepley (Carolina Baseball). Started right away at Liberty where he hit .310/.457/.432 with 13 XBH, but had a breakout summer in the Coastal Plain League to the tune of a .339/.468/.546 slash line with 8 2B, 5 HR, 26 SB and 29 BB to 11 Ks. Carried that momentum over into last season and hit .330/.482/.521 with 12 2B, 9 HR, 25 SB and an impressive 53 BB to just 27 Ks. Kepley then proceeded to have a productive summer on the Cape in which he was named an All-Star. Compact build at 5'8" and 165-pounds with sneaky strength packed into his frame. Crouched stance in the box with a medium-high handset and slightly open front side. Kepley has a minimal load and utilizes a toe tap that leads into a normal stride. With two strikes, Kepley will widen his stance and sit deeper in his base, choke up a bit, and shorten his stride in order to maximize his chances of moving the baseball. Quick hands with present bat speed. Will drop his back side to help create leverage. Overall, it's a compact operation in which he takes an efficient path to contact. Kepley's bat-to-ball skills and hand-eye coordination are outstanding. Last spring at Liberty, he boasted a 90% overall contact rate and a 93% overall IZ contact rate. It was a similar story this summer, as he ran an 89% overall contact rate and a 94% overall IZ contact rate. Kepley's pitch recognition skills and approach are both highly advanced and he rarely expands the strike zone. On the Cape, his overall chase rate was a minuscule 13%. Kepley has high-level barrel skills, a trait that is on display day in and day out. His bat control and adjustability are also advanced. Kepley is capable of using the entire field, but his highest quality of contact does come to the pull side. Hit tool is comfortably a 55, but it's closer to a 60. Kepley without a doubt is a hit-over-power profile, but there is a little bit of thump to the pull side. Has shown the ability to turn on pitches on the inner-half and drive them to the PS. For a player with this kind of profile, he posted respectable Max EVs of 106.1 and 101 during the spring and summer, respectively. Below-average power, but he'll run into a ball on occasion and hit 10+ HRs this spring. Not only is Kepley a plus runner, he most importantly knows HOW to run and get the most out of his legs. Quick first step and really advanced baseball sense both translate to the base paths where he's a chaos causer and speeds up the game for opposing teams. Kepley's speed also allows him to take an extra base on a ball in the gap or down the line. Across 179 games between college and summer ball, Kepley is 89-for-97 on stolen base attempts. Kepley this spring will rove CF for North Carolina. His speed and instincts plays at the position and he has gap-to-gap range with solid closing speed. Gets good reads off the bat. Kepley's arm is fringy, but his overall defensive skillset will give him the opportunity to prove he can stick at the position professionally. Versatile enough to play all 3 OF spots. Kepley is an old-fashioned baseball rat who plays at one speed. Selfless, high-energy player who can affect the game in a myriad of ways. A little reminiscent of Tommy Hawke. Top-5 round type this July.show more