agentconnect-md/agentconnect

AgentConnect – Open‑source multi‑agent platform

What it is – AgentConnect is a self‑hosted, Apache‑2.0‑licensed platform that lets a team run many AI agents side‑by‑side inside the chat and workflow tools they already use (Slack, Discord, Telegram, Google Chat, Lark/Feishu, GitHub, GitLab, Gitea, Linear, …). Agents can be any ACP‑compatible runtime – Claude Code, OpenAI Codex, Grok Build, DeepSeek, Pi, etc. – and they can call each other, keep memory, and act on webhooks, schedules or issue events.

Core pieces

Component	Role
Daemon	Executes agents, holds their workspace and session state, talks directly to the chat platforms and runs scheduled jobs.
Relay (optional)	Central entry point for callbacks/web‑chat; forwards messages to the owning daemon without persisting them.
Control Plane + Web UI	Configuration UI, authentication (Logto), permission management, observability and storage of shared knowledge/skills.

Key capabilities

• Multi‑channel collaboration – agents appear as bots in Slack, Discord, Telegram, Google Chat, Lark, etc., and can also be attached to GitHub/GitLab issues, pull‑requests or Linear tickets.
• Agent routing & selection – uses Jev (a decision‑engine) to decide which agent should answer, which model/runtime to use, and when to invoke other agents.
• Persistent memory & knowledge – each agent has its own memory; reviewed knowledge can be published for any agent to retrieve later.
• Fine‑grained security – per‑agent permissions control which repositories, tools, or other agents it may access.
• Self‑hosted – Docker‑Compose for local dev, Helm chart for Kubernetes, all under Apache‑2.0.

Typical use‑cases

• Collaborative issue triage where a human and several specialist agents (e.g., security, architecture) discuss a ticket in a shared thread.
• Cross‑workspace support: start a conversation in Telegram, bring in engineering on Slack, and return the resolution to the original channel.
• Automated recurring operations triggered by cron or webhook, with human‑in‑the‑loop review.
• Keeping private forks up‑to‑date by having agents monitor upstream repos and propose updates.
• Custom code‑review pipelines that pick reviewers, runtime, and model per pull‑request.

Getting started

1. Clone the repo and run docker compose up -d --pull always – this launches the web console, control plane, relay and a PostgreSQL DB.
2. Open http://localhost:3000, add a daemon, and create your first agent via the UI.
3. For production, deploy the Helm chart (oci://ghcr.io/agentconnect-md/charts/agentconnect).

Community & contribution

• Join the Slack community, star the repo, or share on X/LinkedIn.
• Contributions are welcomed; the project uses Node ≥ 24.12, pnpm 11, and follows a monorepo layout.

Bottom line – AgentConnect is a genuine, open‑source framework for orchestrating multiple LLM‑powered agents across the communication and development tools that teams already use, providing routing, memory, security, and a self‑hosted control plane.