TARDIS

TARDIS

[](https://github.com/mrchevyceleb/TARDIS/actions/workflows/ci.yml) [](LICENSE)

Bigger on the inside. A local-first, always-on multi-agent office for Claude Code, Codex, Grok, GLM, OpenRouter, Fireworks, and local models.

TARDIS is a ship's mind with a crew of Companions: each companion gets a durable home thread, an editable role, model/effort controls, scheduled routines, and a shared team bus. Conversations survive browser changes, service restarts (regenerations), model switches, and rolling context compaction without turning the visible transcript into disposable model state.

[!WARNING] TARDIS can read workspace files and launch powerful local agents. It has **no app-layer authentication**. The server binds to `127.0.0.1` by default. Only expose it through an authentication boundary you trust, such as a locked-down Tailscale tailnet. Never publish port `8091` directly to the internet.

What it includes

• Persistent companions — one forever-thread and editable persona per agent.
• Multi-engine conversations — Claude Code, Codex, xAI Grok, Z.ai GLM, OpenRouter, Fireworks, and local OpenAI-compatible models.
• Agent collaboration — companions can message one another through the ship's built-in team MCP.
• Durable memory — append-only event logs, rolling compaction, restart recovery, and cross-device replay.
• Human-first scheduling — routines defer while a person is actively using a thread and remain draft/review-first for external side effects.
• Workspace desk — browse and edit an ASSISTANT-HUB, open artifacts, pin messages, and use the classic Studio at /studio.
• Voice and PWA support — installable desktop/mobile shell, live voice calls, optional Jarvis wake-word agent, and native Windows workspace links.
• Local operations — health reporting, graceful shutdown