Skip to content

Caelis overview

A local collaboration workspace for models and ACP agents.

Caelis brings models and ACP-compatible agents into one local Session. Participants keep their own conversations, exchange findings, and continue working as you steer the task.

  1. Install Caelis.
  2. Connect a model or agent.
  3. Open a repository and ask for a concrete outcome.
Map this repository and explain how to run its tests.

Your launch directory is the workspace. Sessions live locally; use /resume to return to earlier conversations.

Use /team to configure participant roles. You can follow a participant, send text or images, and coordinate work from the main conversation. See participants.

Local storage and execution are separate from inference: model requests go to your selected provider or external agent. Review approvals and sandboxing before broadening tool access.

Use MCP servers, skills and plugins to bring in tools. Project MCP configuration requires workspace trust. Built-in memory needs no separate installation; you choose whether to configure optional model-assisted processing.

Caelis can also run a single prompt or serve an ACP client. See tools and automation for a first example.

Based on the public project documentationcaelis · 744b6f4 ↗