AgentConnect

AgentConnect is an open-source, self-hostable platform that lets teams deploy and coordinate AI agents across the communication tools they already use — Slack, Telegram, Discord, Lark/Feishu, and GitHub. Licensed under Apache 2.0, the core stack is available to run on your own infrastructure today, while a hosted AgentConnect Cloud console is in waitlist-only early access. The project describes itself as an open-source alternative to Claude Tag, positioning itself for teams rather than individual operators.

What It Is

AgentConnect is a multi-agent coordination layer that sits between your team's communication channels and AI agent runtimes. Instead of each agent living in a developer's terminal, AgentConnect gives every agent a role, a set of tools and permissions, and a presence in shared channels where the whole team can follow the work. Agents can call one another, carry memory between sessions, and be triggered by messages, GitHub events, webhooks, or cron schedules — all managed from a single console.

How the Architecture Works

The stack has three main components:

• Daemon — runs on machines you operate, executes placed agents via the Agent Client Protocol (ACP), owns workspaces and session state, and sends model-provider traffic directly to your provider (never through AgentConnect).
• Relay (optional) — accepts callback-based ingress and webchat, proxies centrally managed MCP and OpenConnector access, and forwards inbound messages to the owning daemon without durable storage.
• Control Plane + Web UI — manages authentication, configuration, placement, permissions, metadata, and observability; stores only explicitly approved organization knowledge and coordination metadata, never message bodies or transcripts.

The daemon-centric design means that if the Control Plane is temporarily unavailable, established sessions and daemon-local schedules continue running.

Open Ecosystem and Integrations

AgentConnect is built around open standards and pluggable components:

• Agent runtimes: Claude Code, Codex, Gemini CLI, DeepSeek, and any ACP-compatible runtime
• Channels: Slack, Telegram, Discord, Lark/Feishu, webchat
• Triggers: GitHub events, generic webhooks, cron schedules
• Memory: AgentConnect-managed, runtime-native, Mem0 (Cloud and OSS), or off
• Tools and apps: Custom MCP providers and OpenConnector-backed services (1,000+ apps cited on the homepage)
• Standards: ACP, MCP, Mem0 OSS, OpenConnector, Logto, OpenTelemetry

Model traffic flows directly between your machines and your provider; AgentConnect never proxies or marks up model usage.

Team Controls and Permissions

AgentConnect adds a team layer that the homepage describes as "the glue every team ends up rebuilding on its own." Each agent gets independently configured runtime, model, workspace, credentials, and daemon placement. Permissions are scoped per agent: which repositories it may access, which tools and skills it may use, which other agents it may call, and which team members can see its sessions. Organization knowledge is reviewed before being shared with agents, and agents can surface new knowledge for team approval.

Update: v1.40.0

The latest release is v1.40.0, published on 2026-08-07, with the repository last pushed on 2026-08-11. The project was created in July 2026 and is under active development, with 33 open issues and 6 forks at the time of indexing. The open-source self-hosted stack is available now via Docker Compose; AgentConnect Cloud (a hosted management console where agents still run in your environment) is in waitlist-only early access. Enterprise deployments with private deployment, SSO, and dedicated support are available by contacting the team.