Most AI tools make you configure half a dozen things before you can do real work. Octomind skips all of that. One binary, one environment variable, and you're running a full AI agent with file editing, shell access, and code search — in under a minute.

Start here.

Install

Octomind ships as a single Rust binary with zero runtime dependencies. Pick your platform:

bash
# One-line install (macOS / Linux)
curl -fsSL https://octomind.run/install.sh | bash

# Or build from source (requires Rust 1.82+)
cargo install octomind

The installer drops the binary into $HOME/.local/bin/octomind. That's it — no Node, no Python, no Docker.

Connect an AI Provider

Octomind talks to 7 providers through a unified provider:model format. The fastest way to start is with OpenRouter, which gives you access to models from Anthropic, OpenAI, Google, and others through a single API key:

bash
export OPENROUTER_API_KEY="your-key-here"

Or connect to a provider directly:

bash
export ANTHROPIC_API_KEY="your-key-here"   # Anthropic
export OPENAI_API_KEY="your-key-here"      # OpenAI
export DEEPSEEK_API_KEY="your-key-here"    # DeepSeek

No config file needed — Octomind picks up environment variables automatically.

Run Your First Session

bash
octomind run

That's it. You're in an interactive session with an AI agent that has full access to your file system, shell, and built-in code search. Ask it to explain your project, refactor a module, write tests — it reads your files, runs commands, and edits code directly.

Want a specialized agent? Use a tap:

bash
octomind run developer:general      # Rust-specific agent
octomind run developer:react     # React frontend agent
octomind run security:owasp      # Security auditor
octomind run lawyer:us           # US legal assistant

Each tap is a pre-built agent configuration: optimized system prompts, scoped tool permissions, and the right model selection for the domain. One command, no setup. For the full mechanics, read how taps work under the hood.

Session Management

Sessions are first-class. Name them, save them, pick up where you left off:

bash
# Start a named session
octomind run --name feature-auth

# Resume it later
octomind run --resume feature-auth

# Or just resume the most recent one
octomind run --resume-recent

When conversations grow long, Octomind's adaptive compression kicks in automatically. It weighs cost against benefit before compressing — only reducing context when it actually saves money. The result: you can work for hours without hitting context limits or context rot, and the agent keeps its critical decisions intact.

Switch Models On the Fly

Inside any session, switch models without restarting:

text
/model anthropic:claude-sonnet-4
/model openai:gpt-4o
/model deepseek:deepseek-chat

Check your token usage and cost anytime with /info, attach screenshots with /image path/to/file.png, or save the session with /save. There are 24 session commands in total — type /help to see them all.

Non-Interactive Mode

Need a quick answer without an interactive session? Pass your message directly:

bash
octomind run developer "Explain the auth module in this project"

Pipe input for automation and CI pipelines:

bash
echo "review this PR" | octomind run --format jsonl

What's Next

You're up and running. Good next steps:

All source code is on GitHub under Apache 2.0. Questions? Open a discussion.