agent6¶

A coding agent with sandboxed commands, resumable sessions, and editable workflows.

agent6 web
Browser interface for desktop or phone

Also available as a streaming CLI and through ACP editor integration.

Installation ¶

Sandboxing supports Linux on x86_64 and aarch64; macOS runs without it, and Windows users should use WSL.

Install from PyPI with uv or pipx, then connect a provider and select a model:

uv tool install agent6      # or pipx install agent6
agent6 connect              # configure a provider
agent6 model default        # choose the required default model

Use Anthropic or any OpenAI-compatible endpoint, including OpenAI, OpenRouter, Ollama, vLLM, llama.cpp, and LM Studio. A provider entry can also use the Responses API (providers.<name>.backend = "responses") or sign in at a configured OAuth issuer. For subscriptions, replace agent6 connect with agent6 connect chatgpt or agent6 connect claude; the latter uses your installed, signed-in Claude Code binary. agent6 completions installs tab completion for your shell.

Usage ¶

A run edits your current checkout. Checkpoint commits go on a separate agent6/<id> branch; checkpointing leaves your current branch's history and index in place. For a separate checkout, use agent6 run --worktree "your task", which starts from committed HEAD without copying uncommitted changes or untracked dependencies.

cd your-repo
agent6 check                # check configuration and sandbox support
agent6 run "your task here"

Commands ask for approval by default. After the task finishes, enter another instruction or type /exit.

Inspect the latest session, then merge its recorded changes when ready:

agent6 sessions show        # status and verification result
agent6 sessions diff        # recorded changes
agent6 sessions merge       # merge into the run's starting branch

The other commands (plan, ask, attach, steer, resume, fork, the sessions verbs, pack and unpack) are on Command line.

To give a finished run more work, use agent6 resume <id> --steer "next task".

Features¶

Sandboxed commands¶

Runs the model's commands in a Linux sandbox with the file and network access you set. Under full or partial isolation, commands cannot read your provider keys. Security model

Verification and review¶

The model can run a configured or inferred verification command; agent6 reports the result or marks the final tree unverified. Reviews the diff with a panel of models you configure and keeps each finding open until you resolve it. Session findings

Resume and fork¶

Saves conversations, tool activity, and Git checkpoints so you can inspect, resume, or fork a session from an earlier turn. Pinned instructions survive compaction; repository memory carries facts across runs. Session recovery

Terminal, browser, and editor¶

The CLI, terminal UI, browser UI, and editor integration share the same engine and session records. The browser also shows the repository's commit graph, diffs, and files with blame. Follow and steer a run

Editable workflows¶

Defines long-running workflows in editable TOML state machines, combining agent runs, tool calls, waits, and branches with offline tests, crash recovery, and replay from recorded results. Agent state machines

Parallel attempts¶

Runs parallel attempts in isolated clones with --parallel N or a list of models, then ranks the results for you to choose which to merge. Parallel runs

Instructions, skills, and MCP¶

Supports repository AGENTS.md instructions, SKILL.md packs, MCP servers, and background commands for development servers and watchers. Skills

Models and budgets¶

The worker, planner, and reviewer can use different models and reasoning effort. Budgets limit spending, tokens for unpriced calls, and supported subscription usage. Budgets

No telemetry unless you configure an export, and no automatic updates.