Zero Setup • Self-Extending • Open Source
Run octomind run doctor:blood and get a fully configured medical AI in 5 seconds.
The community builds the specialists. You just run them.
No config files. No MCP setup. No vendor lock-in.
The AI agent platform that actually works out of the box
octomind run doctor:bloodRun a specialist agent — no config, no setup, no MCP servers to manage
Community builds specialists. You just run them.
octomind run developer:rust
octomind run lawyer:contracts
octomind run finance:analyst
octomind run security:owasp
Community-built specialists. One command: octomind run doctor:blood. All pre-configured.
Agents register MCP servers mid-session. No restart. The agent extends itself.
4-hour sessions stay coherent. Cache-aware cost analysis.
Install and run. Or customize with TOML. Your choice.
Single binary. macOS + Linux. No dependencies.
Everything you need for development, research, and automation:
Register, enable, and disable MCP servers at runtime. Agents extend their own capabilities mid-session.
Spawn specialized sub-agents on demand. Route tasks to the right AI for the job.
Structured multi-step task tracking. Complex implementations with checkpoints.
Smart file reading and editing. Batch operations with atomic multi-line edits.
AST-based code search and refactoring. Structural understanding, not regex.
Full command execution. Background processes. Working directory management.
Switch providers, manage context, track costs — all mid-session:
/model deepseek:v3
Switch providers mid-session. No restart. No lost context.
/model
View current model and available providers
/report
Detailed usage report with cost breakdown
/info
Real-time token usage and costs
octomind run <agent>
Run any specialist from the tap registry
octomind tap <repo>
Add a community or team tap
Work for 4 hours. The agent still knows what it decided in hour one. Cache-aware cost analysis. Pressure-level system. Automatic context management.
Install and run — no config needed. Power users: full control when you need it.
Different models for different tasks. Cheap models for simple work, powerful models for complex reasoning.
Each step can use a different model and toolset. Gather with Gemini Flash, critique with Claude Opus.
Per-request and per-session limits. Never get surprised by a runaway agent.
Landlock (Linux) / Seatbelt (macOS). Lock writes to working directory. Safe execution.
No code required. All defaults in config-templates/default.toml. Override anything.
Agents register new MCP servers at runtime. No restart. No config edit.
Hit a rate limit? Switch mid-session with /model. No restart. No lost context.
| Provider | Format | Features |
|---|---|---|
| OpenRouter | openrouter:provider/model |
Multi-provider access, 100+ models, caching, vision |
| OpenAI | openai:model-name |
Direct API, GPT-4o, O1, vision, tools |
| Anthropic | anthropic:model-name |
Claude 4, vision, tools, caching |
| DeepSeek | deepseek:model-name |
Cost-effective, reasoning models, V3 |
| Moonshot AI | moonshot:model-name |
Kimi K2 series, vision, tools |
| MiniMax | minimax:model-name |
Anthropic-compatible, thinking, tools |
| Z.ai | z:model-name |
GLM models, cost-effective |
google:model-name |
Vertex AI, Gemini 2.0, vision | |
| Amazon | amazon:model-name |
Bedrock, Claude on AWS |
| Cloudflare | cloudflare:model-name |
Edge AI, fast inference, Llama |
| Ollama | ollama:model-name |
Local models, Llama, Mistral, Qwen |
| Local | local:model-name |
Any OpenAI-compatible endpoint |
| Cerebras | cerebras:model-name |
Fastest inference, Llama variants |
Community-built specialists. One command: octomind run doctor:blood. The right model, tools, prompt — all pre-configured.
Agents register new MCP servers at runtime. No restart. No config edit. Self-extending capabilities.
Work for 4 hours. Agent still knows what it decided in hour one. Cache-aware cost analysis.
7 providers. Switch mid-session. Hit a rate limit? Type /model deepseek:v3 and keep going.
Install and run. No setup required. Power users: per-role models, workflows, spending limits, sandbox mode.
Landlock (Linux) / Seatbelt (macOS). Lock writes to working directory. Safe execution.
macOS and Linux. Single binary, no dependencies.
curl -fsSL https://raw.githubusercontent.com/muvon/octomind/master/install.sh | bash
OpenRouter gives you access to all 7 providers with one key.
export OPENROUTER_API_KEY="your_key"
Zero config. The tap registry handles everything.
octomind run doctor:blood
# Start an interactive session
octomind session
# Build from source
cargo install --git https://github.com/muvon/octomind.git
Free and open source software. Use, modify, and distribute freely.
Performance, safety, and reliability. Single binary, macOS + Linux.
Contribute a tap agent — a single TOML file. Domain expertise welcome.