rvaniaaa's banner
rvaniaaa's profile picture

rvaniaaa

@rvaniaaaa • 5,851 subscribers

I break down the AI setups top builders actually use and show you how to copy them. Agents, second brains, cutting AI costs.

Shorts

This is f*cking insane. This tip saved me thousands of dollars. run Opus 5.5, Sonnet 5.5, and Fable 5.1 together, and stop burning Opus on work it was never needed for. the whole idea in one line: the strong model plans, the mid-tier model executes, Fable stays quiet until it's actually needed. roles, broken down: Opus 5.5, high effort, owns the plan and ships the final code Sonnet 5.5, medium effort, splits into explorer (reads the codebase), worker (edits files, runs tests), researcher (pulls docs) Fable 5.1, called through /advisor fable, reads everything happening in the session but stays silent unless something's actually wrong three moments where Fable speaks: → a plan goes out: is this actually the right call? → the same failure shows up again: is the search going nowhere? → the task gets marked finished: did something get skipped? Jev engineering does the same thing one level down. the forks that don't need real thought, which file, which tool, keep going or stop, go straight to Jev and come back in under half a second. the big models only ever see the forks that genuinely need a decision. anyone still running one model for everything is paying Opus prices to decide whether a file exists. drop this into Claude Code: "Rebuild my Claude Code setup around this structure: Look through ~/.claude/agents and .claude/agents for subagents already covering explorer, worker, and researcher. Only create new ones for roles that are missing. Set model: sonnet, effort: medium on each. If an existing subagent is locked to a different model, leave it as is and just list it. In ~/.claude/settings.json, set effortLevel to high and advisorModel to fable. Check for anything disabling the advisor, CLAUDE_CODE_DISABLE_ADVISOR_TOOL, DISABLE_TELEMETRY, or anything blocking feature-flag fetches, plus CLAUDE_CODE_EFFORT_LEVEL, which can override subagent effort settings. Report what you find. Don't change any of it yet. Add one line to ~/.claude/CLAUDE.md: check in with the advisor before a big plan, when the same error shows up twice, and before marking a long task done. Show every change as a diff first. Wait for my go-ahead before touching anything."

This is f*cking insane. This tip saved me thousands of dollars. run Opus 5.5, Sonnet 5.5, and Fable 5.1 together, and stop burning Opus on work it was never needed for. the whole idea in one line: the strong model plans, the mid-tier model executes, Fable stays quiet until it's actually needed. roles, broken down: Opus 5.5, high effort, owns the plan and ships the final code Sonnet 5.5, medium effort, splits into explorer (reads the codebase), worker (edits files, runs tests), researcher (pulls docs) Fable 5.1, called through /advisor fable, reads everything happening in the session but stays silent unless something's actually wrong three moments where Fable speaks: → a plan goes out: is this actually the right call? → the same failure shows up again: is the search going nowhere? → the task gets marked finished: did something get skipped? Jev engineering does the same thing one level down. the forks that don't need real thought, which file, which tool, keep going or stop, go straight to Jev and come back in under half a second. the big models only ever see the forks that genuinely need a decision. anyone still running one model for everything is paying Opus prices to decide whether a file exists. drop this into Claude Code: "Rebuild my Claude Code setup around this structure: Look through ~/.claude/agents and .claude/agents for subagents already covering explorer, worker, and researcher. Only create new ones for roles that are missing. Set model: sonnet, effort: medium on each. If an existing subagent is locked to a different model, leave it as is and just list it. In ~/.claude/settings.json, set effortLevel to high and advisorModel to fable. Check for anything disabling the advisor, CLAUDE_CODE_DISABLE_ADVISOR_TOOL, DISABLE_TELEMETRY, or anything blocking feature-flag fetches, plus CLAUDE_CODE_EFFORT_LEVEL, which can override subagent effort settings. Report what you find. Don't change any of it yet. Add one line to ~/.claude/CLAUDE.md: check in with the advisor before a big plan, when the same error shows up twice, and before marking a long task done. Show every change as a diff first. Wait for my go-ahead before touching anything."

255,654 görüntüleme

STOP STORING NOTES. START COMPILING THEM. a pile of notes isn't memory. it's accumulation wearing memory's clothes. these 7 repos build the other kind, where Claude reads a source once and links it to everything you already have. copy the list: graphify: turns a codebase with its docs, SQL schemas, configs and PDFs into a queryable knowledge graph. 127k stars Understand-Anything: maps a project's code, docs or wiki into a graph you can search and query from your coding assistant. 86k stars claude-obsidian: lets Claude Code build and maintain a Markdown knowledge base in Obsidian, with source-backed notes and grounded answers. 15k stars obsidian-second-brain: gives Claude Code long-term memory as plain markdown notes in an Obsidian vault. 4.8k stars claude-code-memory-setup: an Obsidian vault as persistent memory, graphs for your codebase, and a pipeline that saves insights from your chats. 1k stars second-brain: an Obsidian knowledge base the LLM maintains, built on Karpathy's LLM Wiki pattern. 727 stars llm-wikid: an agent compiles your raw sources into cross-linked wiki pages in Obsidian. 420 stars none of them is your second brain. they're the compiler. the brain is what you feed it pick one and feed it your first 50 sources. before that, a compiled wiki behaves like a plain search engine, so don't quit early. the full setup is in the article

STOP STORING NOTES. START COMPILING THEM. a pile of notes isn't memory. it's accumulation wearing memory's clothes. these 7 repos build the other kind, where Claude reads a source once and links it to everything you already have. copy the list: graphify: turns a codebase with its docs, SQL schemas, configs and PDFs into a queryable knowledge graph. 127k stars Understand-Anything: maps a project's code, docs or wiki into a graph you can search and query from your coding assistant. 86k stars claude-obsidian: lets Claude Code build and maintain a Markdown knowledge base in Obsidian, with source-backed notes and grounded answers. 15k stars obsidian-second-brain: gives Claude Code long-term memory as plain markdown notes in an Obsidian vault. 4.8k stars claude-code-memory-setup: an Obsidian vault as persistent memory, graphs for your codebase, and a pipeline that saves insights from your chats. 1k stars second-brain: an Obsidian knowledge base the LLM maintains, built on Karpathy's LLM Wiki pattern. 727 stars llm-wikid: an agent compiles your raw sources into cross-linked wiki pages in Obsidian. 420 stars none of them is your second brain. they're the compiler. the brain is what you feed it pick one and feed it your first 50 sources. before that, a compiled wiki behaves like a plain search engine, so don't quit early. the full setup is in the article

61,287 görüntüleme

THIS IS F*CKING BRILLIANT. Anthropic’s docs describe one of the smartest ways to run Opus three models work on one task and each does only the part it’s cheapest at. start Claude Code like this: claude -> model sonnet -> advisor opus you stop paying Opus prices for work that doesn’t need Opus > Sonnet runs the session and edits the code > Haiku 5.5 searches the repo and docs, at up to 40x less per token than Opus 5.5 > Opus waits as the advisor and reads the whole session, tool calls included Claude usually calls Opus at three points: > before it locks a plan > when the same error comes back > right before it says it’s done Sonnet drives. Haiku scouts. Opus signs off and if you go one step lower, most of an agent’s day is tiny choices like picking a file or a tool, or deciding whether to retry. Jev takes those first, so Sonnet and Opus only see the moments that are really open the whole setup Sonnet runs the main session Explore on Haiku 5.5 searches and pulls docs, read-only Opus on call as the advisor give Claude Code this setup and the prompt below ↓ “Set up my Claude Code exactly like this: Main model Sonnet, advisor Opus, saved as advisorModel in my settings. One read-only subagent named exactly Explore, on claude-haiku-5-5, to replace the built-in one. One line in CLAUDE.md: ask the advisor before locking a plan, after the same error twice, and before calling a task done, not more often. Check what could turn the advisor off (CLAUDE_CODE_DISABLE_ADVISOR_TOOL, DISABLE_TELEMETRY) and only report it. Show every change as a diff first. Don’t edit anything until I say go.” before you turn it on: every advisor call re-reads the whole transcript with no cache. on a 200k session that’s about $0.80 in input, against $0.04 for a normal request. keep it to a few calls the built-in Explore agent runs on your main model unless a subagent is named exactly Explore. any other name does nothing turning the advisor on or off keeps your prompt cache. switching models rebuilds it it needs the Anthropic API, and it stays off if telemetry is disabled bookmark it before your next long session

THIS IS F*CKING BRILLIANT. Anthropic’s docs describe one of the smartest ways to run Opus three models work on one task and each does only the part it’s cheapest at. start Claude Code like this: claude -> model sonnet -> advisor opus you stop paying Opus prices for work that doesn’t need Opus > Sonnet runs the session and edits the code > Haiku 5.5 searches the repo and docs, at up to 40x less per token than Opus 5.5 > Opus waits as the advisor and reads the whole session, tool calls included Claude usually calls Opus at three points: > before it locks a plan > when the same error comes back > right before it says it’s done Sonnet drives. Haiku scouts. Opus signs off and if you go one step lower, most of an agent’s day is tiny choices like picking a file or a tool, or deciding whether to retry. Jev takes those first, so Sonnet and Opus only see the moments that are really open the whole setup Sonnet runs the main session Explore on Haiku 5.5 searches and pulls docs, read-only Opus on call as the advisor give Claude Code this setup and the prompt below ↓ “Set up my Claude Code exactly like this: Main model Sonnet, advisor Opus, saved as advisorModel in my settings. One read-only subagent named exactly Explore, on claude-haiku-5-5, to replace the built-in one. One line in CLAUDE.md: ask the advisor before locking a plan, after the same error twice, and before calling a task done, not more often. Check what could turn the advisor off (CLAUDE_CODE_DISABLE_ADVISOR_TOOL, DISABLE_TELEMETRY) and only report it. Show every change as a diff first. Don’t edit anything until I say go.” before you turn it on: every advisor call re-reads the whole transcript with no cache. on a 200k session that’s about $0.80 in input, against $0.04 for a normal request. keep it to a few calls the built-in Explore agent runs on your main model unless a subagent is named exactly Explore. any other name does nothing turning the advisor on or off keeps your prompt cache. switching models rebuilds it it needs the Anthropic API, and it stays off if telemetry is disabled bookmark it before your next long session

62,223 görüntüleme

a developer showed me the cleanest multi-agent architecture i've seen. one mission. nobody steps on anyone else. the mission: ship a safe feature. the graph decides what happens next. router reads the mission and asks one question: where should this go? three workers split in parallel. each in their own private work area. researcher finds the evidence. architect designs it. builder creates it. none of them share context. the three workers run independently. everything lands in shared state. facts, decisions, artifacts. one place where the whole picture exists without anyone copying transcripts. then the crew's work converges. integrator combines what was built. reviewer tests quality and safety. human checkpoint approves anything high-impact. then ship. verified output. private context. shared state. that separation is the whole trick. a prompter asks a question. an architect draws a graph. full breakdown with code in the article below.

a developer showed me the cleanest multi-agent architecture i've seen. one mission. nobody steps on anyone else. the mission: ship a safe feature. the graph decides what happens next. router reads the mission and asks one question: where should this go? three workers split in parallel. each in their own private work area. researcher finds the evidence. architect designs it. builder creates it. none of them share context. the three workers run independently. everything lands in shared state. facts, decisions, artifacts. one place where the whole picture exists without anyone copying transcripts. then the crew's work converges. integrator combines what was built. reviewer tests quality and safety. human checkpoint approves anything high-impact. then ship. verified output. private context. shared state. that separation is the whole trick. a prompter asks a question. an architect draws a graph. full breakdown with code in the article below.

27,465 görüntüleme

Videos

rvaniaaaa's profile picture

This is f*cking insane. This tip saved me thousands of dollars. run Opus 5.5, Sonnet 5.5, and Fable 5.1 together, and stop burning Opus on work it was never needed for. the whole idea in one line: the strong model plans, the mid-tier model executes, Fable stays quiet until it's actually needed. roles, broken down: Opus 5.5, high effort, owns the plan and ships the final code Sonnet 5.5, medium effort, splits into explorer (reads the codebase), worker (edits files, runs tests), researcher (pulls docs) Fable 5.1, called through /advisor fable, reads everything happening in the session but stays silent unless something's actually wrong three moments where Fable speaks: → a plan goes out: is this actually the right call? → the same failure shows up again: is the search going nowhere? → the task gets marked finished: did something get skipped? Jev engineering does the same thing one level down. the forks that don't need real thought, which file, which tool, keep going or stop, go straight to Jev and come back in under half a second. the big models only ever see the forks that genuinely need a decision. anyone still running one model for everything is paying Opus prices to decide whether a file exists. drop this into Claude Code: "Rebuild my Claude Code setup around this structure: Look through ~/.claude/agents and .claude/agents for subagents already covering explorer, worker, and researcher. Only create new ones for roles that are missing. Set model: sonnet, effort: medium on each. If an existing subagent is locked to a different model, leave it as is and just list it. In ~/.claude/settings.json, set effortLevel to high and advisorModel to fable. Check for anything disabling the advisor, CLAUDE_CODE_DISABLE_ADVISOR_TOOL, DISABLE_TELEMETRY, or anything blocking feature-flag fetches, plus CLAUDE_CODE_EFFORT_LEVEL, which can override subagent effort settings. Report what you find. Don't change any of it yet. Add one line to ~/.claude/CLAUDE.md: check in with the advisor before a big plan, when the same error shows up twice, and before marking a long task done. Show every change as a diff first. Wait for my go-ahead before touching anything."

rvaniaaa

255,654 görüntüleme • 1 ay önce

rvaniaaaa's profile picture

I still don't understand why everyone is still running agents in a line. I switched to graphs three weeks ago and my fleet finished in the time my single agent used to spend on step two. what slows every agent system I have seen is not intelligence. it is geometry. and almost nobody is talking about it. one engineer used this to rewrite 535,000 lines of code in 11 days. a manual rewrite of that scale could take close to a year. it cost $165,000 in tokens. the graph was not cheap. it was just faster than a human year. a node is one agent with one job. research one competitor. review one file. check one claim. the moment a node owns two independent jobs you lose the ability to parallelize them cleanly, verify them independently, and debug them in isolation. an edge is a dependency. it only exists when data actually moves across it. everything else is a fake edge. a wait you invented that costs time and produces nothing. find the fake edges and the line collapses into something wider. jobs that can run at the same time run at the same time. what used to take the sum of forty steps now finishes in the time of the slowest layer. the pattern behind every serious agent system looks like a diamond. fan out to gather breadth, one agent per angle, all at once. reduce with plain code, no model tokens spent. verify with a fresh skeptic on every finding. synthesize once from what survived. Claude's own research feature uses a very similar pattern in production. the part nobody warns you about: the verifier needs clean context. give it the same conversation the worker had and it is not checking anything. it is nodding along to itself in a different window. a graph of agents sharing one context is a single loop in a costume. it breaks the same way, just later and more expensively. one rule that holds at every scale. a worker and its verifier must never share a context. your agents are not too slow. they are waiting in a line that did not need to exist. full guide in the article. save it before you build your next agent from scratch.

rvaniaaa

259,393 görüntüleme • 1 ay önce

rvaniaaaa's profile picture

Jev just became the fastest-adopted model in AI history. here's what people already built with it 1. jev-ultrafast - the browser agent that picks every click itself, only calling a text model when it actually needs to type something. found real flight results in 7 seconds for $0.0039. 16,758 stars 2. jev-trader - real trading bot placing live limit orders on Monad every 300ms block, judged by Jev alone. 1,911 stars 3. jev-usecases - production security-operations harness where Jev triages incidents and gates every escalation behind a confidence cutoff before anything touches real infrastructure. zero false escalations in the committed test set 4. tax-doc-classifier - sorts real IRS tax forms with 100% strict accuracy across 261 forms, at roughly $0.001 a page 5. jev-drone - a simulated quadrotor clears a five-station obstacle course by camera alone, Jev judging the situation twice a second 6. killmyidea - describe your startup idea, Jev scores it from every angle, then hands back kill, fix, or ship in seconds, not days 7. jev-curate - streams Parquet and JSONL rows through typed judgments at 1,500+ rows a second, keeping only what clears the bar 8. pg-jev - a PostgreSQL extension that lets you ask your own database tables plain-English questions and get a real answer back, no SQL required eight repos. zero generated words. every single one returns a typed answer against a question someone already defined full setup below, then run the three-question test from the article before you build a ninth

rvaniaaa

86,531 görüntüleme • 1 ay önce

rvaniaaaa's profile picture

I still don’t understand why everyone isn’t running one of these yet. mine has been compiling for six months and it already knows things I’ve forgotten I knew Andrej Karpathy published a folder structure in April 2026. 5,000 stars. 16 million views. engineers couldn’t stop sharing it. not a model, not a tool, a folder structure. the sentence that explains why: “RAG re-derives knowledge on every query. a compiled wiki derives it once and keeps it current” most people who try building one hit a wall around month three anyway. the notes pile up, the links exist, and then it just sits there. you’re still the one doing all the remembering here’s the sentence that explains that wall too: keeping fifty interlinked notes current, consistent, and cross-referenced is work no human sustains. so people stop maintaining it, and it turns into a graveyard of good intentions the fix isn’t discipline. it’s making the system do the maintenance instead of you here’s the gist drop any source into a folder called raw, untouched, ground truth Claude reads it once, writes a structured page into wiki, links it to everything already there, flags anything that contradicts what you already know one file called CLAUDE.md sits at the center holding who you are, so you stop re-explaining yourself every session ask it anything across everything you’ve ever fed it, and it answers from months of compiled understanding instead of starting from zero that loop is the whole system. Karpathy’s repo got 16 million views because engineers finally saw the difference written down: retrieval answers questions, compilation builds understanding five minutes to set up. you’ll never open an empty chat and re-explain your life again full build, every step, in the article. save it now

rvaniaaa

60,398 görüntüleme • 2 ay önce

rvaniaaaa's profile picture

install these before your next long Claude Code session if your Opus 5.5 bill hurts and you don’t know where to start, copy this: claude-code-router: sits in front of Claude Code and sends each request to the cheapest model that can do it. Haiku does the grep, Opus only gets the hard part. 37k stars rtk: shrinks test, build and git output before the model sees it. one binary, up to 90% less bash output in your context. 83k stars claude-mem: remembers what happened in your past sessions and feeds back only the relevant part, so you stop paying to re-explain your project every morning. 100k stars context-mode: a huge tool output goes into a local database instead of your context window, and Claude pulls out only the lines it needs. 18k stars caveman: makes Claude answer in short telegraphic sentences. code stays identical, and output is the expensive side on Opus 5.5 ($20 per million). 51k stars token-optimizer: a local dashboard that shows where your tokens actually went, and keeps your work alive across compactions. 1.9k stars token-savior: replaces whole-file reads with symbol lookups, so Claude reads the function and not the 2,000 lines around it code-review-graph: maps your repo as a graph and gives Claude only the code connected to the change you’re reviewing jev-ultrafast: a browser agent where Jev picks every click, and a text model wakes up only when something has to be typed. 16,758 stars none of them makes Opus smarter. every one stops it from paying for work it never needed to do the full setup with scripts is in the article. do the measuring step first, so you know what you actually saved

rvaniaaa

25,052 görüntüleme • 1 ay önce

rvaniaaaa's profile picture

STOP BURNING YOUR OPUS 5.5 LIMIT FOR NOTHING these 8 repos cut what Claude reads, writes and re-reads. copy the list before your next long session: headroom: shrinks tool output, logs and files before they reach the model, and keeps the originals so Claude can pull them back when it needs them. 75k stars claude-context: semantic search over your whole codebase, so Claude fetches the relevant snippets and doesn’t read whole files to find them. 12k stars lean-ctx: a local layer that compresses file reads and shell output. a file Claude already read costs about 13 tokens the second time. 3.8k stars openwolf: hooks that keep a project index and session notes, so Claude skips files it already read and doesn’t lose its work at compaction. 2.1k stars claude-token-efficient: one CLAUDE.md that keeps answers short. nothing to install, and output is the $20-per-million side of Opus 5.5. 6.1k stars jcodemunch-mcp: indexes the repo once, then Claude asks for one function and gets just that. it claims 95% less code reading. 2.5k stars ccusage: reads your local Claude Code logs and shows tokens and cost by day and by session. 19k stars Claude-Code-Usage-Monitor: live usage bars and a forecast of when you’ll hit your limit. 8.6k stars none of them makes Opus better. each one just stops it from paying for tokens it didn’t need start with the last two: see where your limit actually goes, then pick two tools from the top. the full setup with scripts is in the article

rvaniaaa

12,381 görüntüleme • 1 ay önce

Daha fazla içerik yok.