Get Started
Quick Start
After install, jump into a project and run minnal. That is the whole onboarding — but here is what to do next.
Launch the TUI
cd path/to/your/project minnal
On first launch you will be asked to connect a model provider. Use the /connect slash command to log in to Anthropic, OpenAI, Gemini, Copilot, or paste an API key for any other provider.
Initialize the project
Run /init inside a fresh repo. minnal scans the codebase, infers the conventions, and writes an AGENTS.md file at the root that future sessions read automatically.
/init
Ask and build
Once connected, talk to it like a sharp junior engineer:
How is auth wired in @src/api/index.ts? Add a /logout route that revokes the refresh token, then write a unit test.
Non-interactive runs
minnal run "summarize the last 10 commits in this repo"
Use minnal run in scripts, CI jobs, or git hooks. Output is plain text by default; pass --json for machine-readable output.
Resume a session
Every session gets a memorable name (fox, oak, vega, …). Resume by name from anywhere:
minnal --resume fox
You can also resume sessions from other harnesses — Claude Code, Codex, OpenCode, and pi. minnal will pick up the conversation transcript and continue.
Server mode
# Start the persistent core minnal serve # Attach a TUI client minnal connect # Attach a second client to the same server minnal connect
The server keeps memory, sessions, and skill state hot across many attached clients — including remote ones over Tailscale.
Dictation
minnal dictate
Voice input via your configured speech-to-text command. minnal does not bundle an STT engine — bring your own (whisper.cpp, macOS say, etc.).
Next steps
- Configure providers in detail — see
Providers & OAuth. - Wire up MCP servers and skills.
- Try Swarm: spin up two minnals in the same repo.