Install agents,
not frameworks.
Open-source runtime for specialist AI agents
Stop wiring MCP servers, glue code, and credentials. One command gives you a specialist with the right model, tools, and prompts — ready to run.
octomind run assistant:conciergeWhy Octomind?
The four pains every agent builder hit in 2026 — and what Octomind does instead
The Problem
- ✕ Config Wars Stitching three tools, writing glue code nobody wants to own. No central registry, no quality signal.
- ✕ Generic AI fails in expert domains Wrong drug dosages. Hallucinated case citations. NY moving to ban AI chatbots posing as lawyers.
- ✕ Sessions break at hour 4 Naive truncation drops the decisions you need. Quality collapses. Restart.
- ✕ Bills surprise you $7K daily overages. No per-task budget. No kill switch.
Octomind
- → Tap registry One command installs a specialist with the right model, MCP servers, and prompts — pre-wired by domain experts.
- → 117 specialists across 28 domains Lawyer (9 jurisdictions), doctor, engineer, devops, finance, security, content, launch. Grounded MCP, not vibes.
- → Adaptive compression Cache-aware, structurally preserving. 4-hour sessions stay sharp. Smaller context, faster responses, lower cost.
- → Hard spending caps Per-request and per-session enforced — agent stops, falls back, or warns before the bill.
The harness matters as much as the model
25 tasks harvested from pull requests merged in 2026 — python, php, rust, c++, js — graded by each project's own held-out tests. Four agents, stock settings, no tuning.
| Agent | Model | Solved | Judge Σ / 2500 | Cost | Wall time |
|---|---|---|---|---|---|
| octomind | glm-5.2 (open) | 24/25 | 2264 | $63.43 | 3.6h |
| claude code | claude-opus-5 | 23/25 | 2262 | $81.79 | 6.7h |
| codex | gpt-5.6-sol | 21/25 | 2127 | $14.86 | 1.0h |
| opencode | glm-5.2 (open) | 19/25 | 2093 | $129.54 | 3.3h |
- Same model, same endpoint as opencode — 24 vs 19 solved at half the cost. The difference is pure harness: context discipline and supervision.
- octomind's cost is its worst case. glm-5.2 ran via Ollama cloud with no prompt caching — every token at list price — while Opus billed ~97% of its context re-reads at 1/10 cache rates, and still came out behind on solves, cost, and time.
- Every task is a real merged fix — mostly landed after model training cutoffs — proven fail-to-pass before benching, validated by hidden maintainer-written tests plus an LLM judge.
One Command. Any Specialist.
Like Homebrew for AI agents — community-built specialists you run with a single command
octomind run doctor:blood"Interpret these blood test results for a 45-year-old male"octomind run devops:kubernetes"Why is my pod stuck in CrashLoopBackOff?"Browse 117 specialists in the Tap registry. Or build and share your own.
octomind run developer:generaloctomind run lawyer:contractsoctomind run finance:analystoctomind run security:owaspOne command runs a whole team
A workflow chains specialists into a pipeline: research hands the writer a brief, an auditor scores the result, an editor fixes it — and the run halts rather than making things up. Pipe something in; a finished piece comes out.
octomind workflow promote.tomlpromote turns one article into platform-native social drafts — a researcher grounds
it, a writer drafts, and the polish loop re-edits until the audit signs off.
Question to Verified Cited Report
Investigate a question from background, evidence, and counter-argument angles in parallel, synthesize one cited report, then loop a separate groundedness judge (claims checked against their sources) until GROUNDED, capped at 2 iterations.
launchIdea to Go-to-Market Kit
Take a product idea from market exploration to an honest go/no-go, then — only if viable — produce brand, pitch, ads, and a zero-budget launch plan.
deep-reviewParallel Multi-Lens Code Review with Adversarial Verify
Review a change (branch diff, current changes, or a path) from independent angles in parallel — correctness, security, concurrency, performance, design/tests — then adversarially verify every finding against the real code (refute-first, because consensus is not correctness), dedup, and synthesize one severity-ordered review with a verdict.
Built for Customizable AI Agents
No-code agent runtime that doesn't box you in
Stop Context Rot
Adaptive compression saves 72.5% of tokens automatically. Work for 4+ hours — the agent remembers decisions from hour one. Cache-aware cost analysis ensures compression only triggers when it saves money.
20+ Providers, Zero Lock-in
OpenRouter, OpenAI, Anthropic, DeepSeek, Google, Ollama, and more. Switch models mid-session with /model. Hit a rate limit? Swap providers instantly — no restart, no lost context.
Dynamic MCP
Agents extend themselves at runtime. Register new MCP servers mid-session — no restart, no config edits. The agent decides what tools it needs and loads them on the fly.
Zero Config, Full Control
Single Rust binary — installs in 30 seconds. Works out of the box with sensible defaults. Power users: customize everything with TOML — per-role models, spending limits, sandboxed execution.
Nothing to install — run it in the cloud
Octomind Cloud is open to everyone: your agents on our machines, models included, zero API keys, sessions you can pick up from any device. Free tier, no card required.
Get Started in 30 Seconds
Three steps. No config files. No API keys to collect. Just run.
Install
brew install muvon/tap/octomind Install via Homebrew on macOS
Connect Models
octomind login One login connects the Hub — free models included, no provider accounts. Or bring your own keys: OpenRouter, OpenAI, Anthropic, Ollama, and 20+ providers.
Run
octomind run developer:general Pick any specialist. Zero config required.
Alternative Install Methods
curl -fsSL https://octomind.run/install.sh | bashcargo install octomindgit clone https://github.com/muvon/octomind
cd octomind && cargo build --release100% Open Source
Octomind is fully open source — no vendor lock-in, no hidden costs. Read every line of code, self-host on your infrastructure, and build agents that are actually yours.