AgentConnect is a platform where teams and several AI agents work together in the chat and code tools the team already uses: Slack, Telegram, Discord, Lark, Google Chat, GitHub, GitLab, Gitea, and Linear. You give each agent a role, a runtime, a model, a workspace, tools, and a machine, then @ it in a conversation, issue, or pull request. Agents can call one another, keep their own memory and skills, and start work from a message, an issue, a pull request, a webhook, or a schedule. It positions itself as an open, multi-agent alternative to Claude Tag.
Agents run on Claude Code, Codex, Grok Build, DeepSeek, Pi, or any other runtime that speaks the Agent Client Protocol (ACP), and different runtimes run side by side in the same workflow. Reusable Decisions powered by Jev decide when agents respond, route new conversations to specialist agents, and pick a runtime and model for each new session. Reviewed organization Knowledge is published once and found by every agent on demand, and permissions set who can see each agent and session, which repositories and tools it may use, and which other agents it may call.
The architecture has three parts. Daemons run the agents over ACP, own workspaces, session state, platform connections, and schedules, and send model traffic directly to providers. An optional relay accepts webhook ingress and webchat and proxies centrally managed MCP access. The control plane and web console handle authentication, configuration, placement, permissions, and observability, and store coordination metadata rather than message bodies, so established sessions and local schedules keep running if the control plane is briefly unavailable. The stack is self-hosted with Docker Compose or a Helm chart on Kubernetes, and a hosted cloud version exists.
Typical uses include shared issue triage, support that spans trusted workspaces, recurring operations started from schedules or webhooks, keeping private forks current with upstream, and customized code review with general, architecture, or security reviewers.
Features
- Multi-agent teams: agents with distinct roles that call on each other inside shared conversations
- Chat and code platforms: Slack, Telegram, Discord, Lark, Google Chat, GitHub, GitLab, Gitea, and Linear
- Any ACP runtime: Claude Code, Codex, Grok Build, DeepSeek, Pi, and other ACP-compatible agents side by side
- Per-agent configuration: runtime, model, workspace, tools, and machine set independently for each agent
- Decisions: Jev-powered rules for when agents respond, which specialist gets a conversation, and which runtime and model a session uses
- Memory and Knowledge: per-agent memory and skills, plus reviewed organization Knowledge every agent can search
- Triggers: work starts from a message, issue, pull request, webhook, or schedule
- Permissions: control who sees each agent and session, which repositories and tools it uses, and which agents it may call
- Fleet console: one web view of agents, sessions, schedules, tools and skills, knowledge, and daemons
- Daemon execution: agent execution and workspaces stay on daemons in the environment you operate
- Self-hosting: Docker Compose or a Helm chart for Kubernetes, with a hosted cloud option
