AgentConnect places AI coding agents inside the conversations and workflows a team already uses. You mention an agent in Slack, Telegram, Discord, or Lark, or attach it to repositories, issues, and pull requests on GitHub, GitLab, and Linear, and it works there where colleagues can follow along. Any runtime that speaks the Agent Client Protocol (ACP) can be used, including Claude Code, Codex, Grok Build, DeepSeek, and Pi.
Each agent is configured separately: its runtime, model, workspace, tools, machine, and sandbox policy. Agents can call one another, so a general code reviewer can bring in an architecture or security reviewer only when a change needs it. Work starts from a message, an issue, a pull request, a webhook, or a schedule, and each agent carries its own memory and skills plus reviewed organization-wide Knowledge that any agent can look up.
Permissions are explicit: you decide who can see each agent and session, which repositories and tools it may use, and which other agents it may call. Documented use cases include shared issue triage, support that spans trusted workspaces, recurring scheduled or webhook-driven operations, keeping private forks current with upstream changes, and customized code review.
The deployment has three parts. A daemon runs the placed agents over ACP, owns workspaces and session state, holds the platform connections and schedules, and sends model-provider traffic directly. An optional relay accepts callback ingress and webchat and proxies centrally managed MCP access without durable storage. A control plane with a web console handles authentication, configuration, placement, permissions, metadata, and observability, storing coordination metadata rather than message bodies or session streams; when it is unavailable, established sessions and daemon-local schedules continue and configuration changes resume after reconnection.
You can run the stack yourself with Docker Compose or on Kubernetes through the Helm chart published under each release version, and a setup server on loopback configures browser authentication, the deployment's GitHub, Slack, Google, and Lark apps, and the sign-in methods you offer.
Features
- Any ACP runtime: Claude Code, Codex, Grok Build, DeepSeek, Pi, and other ACP-compatible agents run side by side
- Chat and code surfaces: agents attach to Slack, Telegram, Discord, and Lark bots, and to GitHub, GitLab, and Linear
- Agent-to-agent calls: agents with different roles hand work to each other while people follow the thread
- Per-agent configuration: runtime, model, workspace, tools, machine, and sandbox policy are set independently
- Memory and Knowledge: each agent keeps its own memory and skills, with reviewed organization Knowledge available on demand
- Scoped permissions: visibility, repository and tool access, and which agents may be called are all set per agent
- Triggered work: sessions start from a message, issue, pull request, webhook, or schedule
- Self-hosting: an Apache-2.0 stack you run yourself, keeping agent execution and workspaces in your own environment
- Console: one place for agents, sessions, schedules, tools, knowledge, and daemons
