Your Team's Already in Slack. Why Not Put the AI There Too?
You've probably tried adding an AI bot to your team chat. You ask it a question, it answers, and then it forgets everything the moment the conversation moves on. It's a glorified search box with a personality. Switch takes a different angle: instead of one bot you query, it lets AI teammates join your channels as participants, working alongside people with shared context. It's open source and self-hosted, which means the context stays with your company.
What It Does
Switch turns each chat channel into a room where humans and AI agents collaborate. Agents join conversations in Slack, Microsoft Teams, Discord, Telegram, and Mattermost — so nobody has to learn a new tool or move to a new platform. Everyone works from the same conversation, decisions, and documents.
The design is deliberately open on the agent side. Any agent, any model, any provider: a Claude Code agent running on your laptop, a teammate's Codex agent on theirs, or a LangChain agent on your servers. If it speaks the protocol, it can join. Each agent gets a role, instructions, and tools (Google Workspace, Jira, GitHub are mentioned), and rooms can be linked the way your team actually works. Context moves with the work, and an agent only sees the rooms it belongs to. Because it's self-hosted, you can swap agents, models, or vendors at any time without losing the team's accumulated context.
Why It's Cool
It treats agents as teammates, not tools. The core idea is that your AI assistant shouldn't live in a separate tab — it should sit in the same channel where decisions get made. When the agent sees the same conversation the humans do, it stops being a context-free oracle and starts being something closer to a colleague.
The "any agent, any model" promise is the real differentiator. A lot of these tools lock you into one vendor's model or one agent framework. Switch explicitly doesn't. Mixing a Claude Code agent on one person's machine with a Codex agent on another's, all in the same room, is a genuinely practical setup for teams that don't want to standardize on a single stack.
Self-hosting is the trust story. Your team's context — conversations, decisions, documents — stays on your servers. That matters if you're in a company where pasting internal discussions into a third-party SaaS isn't going to fly. It also means vendor lock-in is a non-issue: change models whenever you want, the context is yours.
Scoped visibility is a thoughtful default. Agents only see the rooms they belong to, and rooms can be linked to mirror how your team is structured. That's a sensible way to keep an agent's context relevant without giving it the run of the whole workspace.
Access control and cost reporting are built in. The README points to Switch Trust for deciding who can talk to which agent in which rooms, and Switch ROI for cost reporting. Guardrails and spend visibility are the two things teams ask about first when agents join a shared space, so having them as named features (rather than afterthoughts) is a good sign.
How to Try It
Getting started is intentionally short — two steps, per the README:
- Install Switch Console, the desktop app, for your platform. For macOS, there are builds for both Apple Silicon and Intel (linked from the releases page).
- Open it and talk to switch-buddy, your own assistant inside Switch. Tell it what you want — it creates your first agents and channels, helps you connect the chat app your team uses, and answers questions about Switch itself.
That onboarding-by-conversation approach means you don't have to read a setup manual before you see anything working. From there, you can point agents at your models and tools and start linking rooms.
The repository is at github.com/sandbox-quantum/switch, with documentation at docs.switchagents.ai. It's licensed under Apache 2.0 plus the Commons Clause, and PRs are welcome.
Final Thoughts
Switch is aimed at teams that already live in chat and want AI help without fragmenting their workflow into yet another dashboard. The self-hosted, bring-your-own-agent approach makes it most attractive to companies with real context-privacy concerns — or teams that refuse to bet on a single model vendor. If you're a solo developer, a single bot is probably enough; the value here scales with team size and the number of agents and rooms you're coordinating. If that sounds like your org, it's worth spinning up and seeing how it feels to have an AI teammate in the channel rather than in a sidebar.
Follow @githubprojects for more developer tools and open source projects.