Launching today
Caelis
Open-source runtime for long-running AI agent work
1 follower
Open-source runtime for long-running AI agent work
1 follower
Caelis is an open-source, terminal-first runtime for long-running AI agent work. Run GPT-6 Astra as the lead agent and bring specialist models and ACP agents into the same local session. Participants exchange messages, use tools, and keep their own conversations. Follow their work in split panes, steer agents directly, and resume durable sessions with built-in memory and approval controls.
Hey Product Hunt 👋 I'm YongZhi, building Caelis.
I wanted AI agents to stay with an engineering task: inspect a repository, split the work, exchange findings, run tools, and carry useful context into the next session.
For the GPT-6 Astra Challenge, Astra acts as the lead agent in Caelis while specialist models and ACP-compatible agents collaborate in the same local session. Astra plans and integrates the work; Caelis provides participant messaging, tools, approvals, durable sessions, and built-in memory.
What you can try today:
• Coordinate repository exploration, implementation, testing, and review with multiple participants.
• Open a participant in a split pane, follow its conversation, and send it input directly.
• Use /resume to return to earlier sessions, with Remember and Recall for durable memory.
• Bring your preferred model providers, ACP agents, MCP tools, and skills.
Get started on macOS, Linux, or Windows:
1. Install with npm, PowerShell, or the shell command in the README linked below.
2. Run caelis inside a repository, then use /connect and /model to choose a provider and model.
3. Try: “Review this change with two participants: one for correctness and one for test coverage. Have them exchange findings, then summarize the issues. Do not edit files.”
Caelis is free and open source under Apache-2.0. Bring your own model access; provider charges or subscription limits still apply. External ACP agents support different capabilities.
I'd love feedback on collaboration, long-running task visibility, and the first-run experience. What engineering task would you put a team of agents on?
Website: https://caelis.dev/
Source, releases, and setup: https://github.com/caelis-labs/c...