# BattleLab (agent-sessions)

> A self-hosted, mobile-first command deck for AI coding agents — Claude Code, Codex, opencode,
> Gemini, Antigravity, and Kimi Code. You are the Commander: every agent runs in its own persistent
> session, HQ briefs you on what needs you through a General you pick, and Missions (beta) take an
> objective and run it. Free and open source (AGPL-3.0-or-later).

BattleLab is the public name for the `agent-sessions` project. It is self-hosted — it runs on
your own Linux host and BattleLab keeps your sessions on your machine (the coding agents still reach their own providers; HQ, reviews and Missions talk only to the AI endpoint you configure). Install with one line:
`curl -fsSL https://battlelabos.com/install.sh | sh`.

## Key facts
- Supported AI coding agents: Claude Code, Codex, opencode, Gemini, Antigravity, and Kimi Code; Claude, Codex and opencode also run as API sessions (structured view, approvals, message queue).
- License: AGPL-3.0-or-later (free and open source). The BattleLab name and logo are trademarks and are not licensed with the code.
- Install: one-line rootless installer; atomic releases + one-step rollback + opt-in self-update. A first-run setup covers optional 2FA, the detected agents, the AI endpoint and — once AI is connected — the HQ contact.
- HQ: ask what needs you, have it start or nudge sessions, and follow up on what it started. Answers come from the primary contact — one of thirteen fictional Generals, a different one each day, or none.
- Dashboard start page: what needs you, running sessions, missions, recent work and per-agent quota.
- Missions (beta): an objective in; a proposed plan (project, agent, objectives that define done), run on the operator's go, objectives settled on evidence. Beta — still settling.
- AI: one OpenAI-compatible endpoint (local servers such as Ollama, LM Studio, llama.cpp, vLLM, or a hosted API) powers HQ, session reviews, auto-sort and Missions; what those features send goes only to that endpoint (a local model keeps it on the machine, a hosted API receives it); the coding agents reach their own providers independently. The key is only sent to the host it was saved for.
- Persistent terminal sessions via dtach PTYs; clean console-style scroll-up history.
- Files & git: every session has a Files panel rooted at its working directory (upload, edit in place) and a Git tab (branch, ahead/behind, changed paths, diffs; fetch, pull, stage, commit, push — refusing rather than forcing when the tree moved).
- Library: playbooks, prompt templates, variables, project memory, automations and skills.
- Tactical map of sessions grouped into projects; AI auto-sort assigns new sessions to the right project.
- Mobile-first web terminal (compose bar, control keys, dictation, image paste), push notifications, QR sign-in.
- Home Free: stream the deck to any browser through a blind, end-to-end-encrypted relay; 4-hour sessions, no account.
- Single-admin tool; run it behind a reverse proxy (TLS + auth) or through the relay; optional TOTP two-factor auth.
- Full documentation at https://docs.battlelabos.com/ — one page per feature, the trust model and enforced security properties, plus the CLI/API/environment reference.

## Links
- Home and install: https://battlelabos.com/
- Documentation (every feature + the full security model): https://docs.battlelabos.com/
- Created by: https://superstatus.io