agent6¶

A coding agent that jails model commands and uses editable state machines for long-running tasks.

The model can write code and ask to run commands, but those commands go through a jail with restricted filesystem and network access. Long-running workflows can be written, reviewed, edited, resumed, and replayed as declarative state machines instead of being left to an open-ended agent loop.

Per-command sandbox¶

Each command the model runs is jailed on its own with Landlock + seccomp; the default strict isolation adds user namespaces + pivot_root, a read-only .git, and no network for jailed commands. Not one coarse container around everything.

Open-weight or frontier¶

Works with Anthropic and any OpenAI-compatible endpoint (OpenAI, OpenRouter, Ollama, vLLM, llama.cpp, LM Studio), and the prompts and tools are tuned to stay usable on cheaper open-weight models.

Resumable and forkable¶

State is snapshotted before every model call and committed after every passing step. Resume an interrupted run from its snapshot, or fork a new run from any past turn while the original stays intact.

Plan, run, review, ask¶

Edit-free planning and repository Q&A, a diff-review panel, and the run loop, each its own command. A live terminal dashboard, full transcripts, and searchable run history.

State machines¶

agent6 machine composes longer automated tasks from runs, sandboxed tool calls, waits, and branches: drafted by the model, reviewed by you, journaled, and replayable.

Small, fixed tool surface¶

The model's tools are a fixed set declared in one file. The only way to add more is an operator-configured MCP server, off by default. No telemetry, no auto-update.

The terminal UI¶

agent6 run is headless by default: the run's conversation streaming in your terminal, the CLI mode. agent6 tui opens the hub instead: every run for the repository, with its mode, status, and cost, where you open a session to read its live conversation, toggle the dashboard (Ctrl+D), or scroll the event log. agent6 run --tui jumps straight to that conversation view; -i drives the run from a stdin REPL. The tour has a still of each screen.

The web UI¶

agent6 web serves the same views in a browser: start a run and watch it stream, steer it, approve prompts, answer questions, read the transcript, and browse and run state machines, from a desktop or a phone. It binds 127.0.0.1; put tailscale serve in front for encrypted remote access. See the web UI.

Install¶

uv tool install agent6        # or: pipx install agent6

agent6 needs Linux for the sandbox, Python 3.12 or newer, and an API key for at least one provider. macOS runs unsandboxed behind a startup warning; on Windows use WSL. See installation for the full requirements.

Run¶

agent6 connect                       # pick a provider, paste an API key (once)
agent6 model worker anthropic claude-sonnet-4-6

cd your-repo
agent6 run "add a --json output mode to the CLI"

agent6 infers a verify command when you have not set one, commits each step that passes it, and stops when the run finishes or a budget ceiling is hit. The getting started guide covers the first run and recovering one that went wrong.