AI Briefing
KO

Agent Switch Released

·2026.08.28 12:30

Key point

Agent Switch, a self-hosted infrastructure that connects AI agents to existing messengers like Slack and Teams for collaboration with humans, has been released.

1 / 2

Details

Agent Switch is a self-hosted collaboration infrastructure that allows humans and AI agents to work together within existing collaboration tools such as Slack, Microsoft Teams, Discord, Telegram, and Mattermost without replacing them. The project does not provide agents or models directly; instead, it offers a layer for managing channel-specific instructions, roles, permissions, connected knowledge, and tasks. It focuses on connecting agents already in use, such as Claude Code, OpenAI Codex, and OpenCode.

Architecture and How It Works

Agent Switch Core operates around Tuwunel, a Matrix homeserver. Humans converse in their existing messengers, while agents join Matrix rooms via the Agent Bridge. The Collaboration Bridge connects platform-specific transport methods (such as Slack Socket Mode and Teams HTTP listeners) to map channels to Matrix rooms. Agents send messages via HTTP, and the server pushes events using Server-Sent Events (SSE). State management uses PostgreSQL, and operators control the management layer via a dashboard.

Installation and Operational Considerations

Installation is possible on macOS, Windows, and Linux via the desktop app Switch Console. Windows and Linux builds are currently in early access, requiring verification of behavioral differences per release before production deployment. Codex users can install connectors via the plugin marketplace, but since generated credentials are stored in plaintext, adding them to .gitignore and checking access permissions is essential. Running continuously on a remote host requires Docker, and there is an operational burden of manually restarting the server and agents after reboots.

License and Suitability

Released under the Apache 2.0 license with modified Commons Clause conditions, which restrict selling paid services or hosting that derive substantial value. It is suitable for teams that want to maintain their existing messengers and agent tools while separating roles and contexts for operations. Conversely, it may be difficult to adopt immediately for teams that need to provide agents themselves or lack the capacity for complex credential management.

This summary was generated automatically by AI. Check the original for the author's claims and context. Copyright belongs to the original author.

Our guide explains how the AI works. Report summary errors, attribution issues, or removal requests via Contact.