Switch AI Team Workspace
Bring AI teammates into your existing chat channels so people and agents can work from shared context.
- Source repo
- sandbox-quantum/switch
- Stars
- ★ 3.8k
- Last updated
- today
- License
- NOASSERTION
- Primary language
- Python
- FA score
- 41/100 · Major gaps
At a glance
- How it runs
- Works with
- Universal · cross-platformCodex · Claude Code (Partial support)
- Setup effort
- High · needs real infrastructure
- You'll need
- Typical use
- A development team wants colleagues and a Claude Code agent to work on a feature in one Slack channel with the discussion trail intact.
- Not a fit if
- Teams seeking only a hosted chat bot
- Teams unwilling to enable opt-out anonymous telemetry by default
- Source review
- 41/100 · Major gaps
What does this agent do, and when should you use it?
Switch turns channels in Slack, Microsoft Teams, Discord, Telegram, and Mattermost into rooms where people and AI teammates work together. Teams can self-host Switch and connect agents and models that speak its protocol; the README names Claude Code, Codex, and LangChain agents as examples. Each agent can have a role, instructions, and tools, and can see only the rooms it belongs to. Switch does not provide chat apps, agents, or models; it connects them to team workflows. The quickstart starts with the Switch Console desktop app, where switch-buddy helps create agents and channels and connect a team chat app; the README also documents remote hosting.
A user installs Switch Console and talks to switch-buddy, which helps create initial agents and channels, connect a chat app, and answer questions about Switch. Teams can configure agent roles, instructions, and tools such as Google Workspace, Jira, and GitHub, and link rooms around their workflows. Agents and people share conversations, decisions, and documents in chat channels; README examples include an agent opening an incident channel, collecting an error rate, and identifying a related deploy, as well as agents triaging feature requests or bug reports. An agent sees only its assigned rooms. Teams can run Switch on their own servers; Console telemetry is on by default and can be turned off, while server telemetry is enabled only with TELEMETRY_ENABLED=true.
- A development team wants colleagues and a Claude Code agent to work on a feature in one Slack channel with the discussion trail intact.
- An engineering lead wants a Codex agent focused on one system area so colleagues can ask it questions in the relevant channel.
- A product operations team wants an agent to gather details in a feature request channel and file requests in Jira, Confluence, or Notion.
- A support team wants an agent to answer from runbooks and correct a runbook in its owning channel when it finds an error.
- An on-call team wants people and agents to handle alerts in a shared channel and capture incident knowledge afterward.
How do you install or deploy this agent?
The README quickstart begins by installing Switch Console. It lists macOS builds for Apple Silicon and Intel, a Windows installer, and AppImage and DEB builds for Linux x64 and ARM64. Open the app after installation and talk to switch-buddy to start creating agents and channels. The README links to a remote hosting guide, but the supplied material does not include server deployment commands or a complete dependency list.
How do you use this agent?
Install and open Switch Console, then tell switch-buddy what agents, channels, and workflows you want. It helps create the first agents and channels, connect the team's chat app, and answer Switch questions. Connect agents such as the README's Claude Code or Codex examples; the documentation can also be accessed through an MCP client.
What are this agent's strengths and limitations?
- Works with five listed team chat platforms: Slack, Microsoft Teams, Discord, Telegram, and Mattermost.
- Supports self-hosting and agents and models that speak its protocol; the README explicitly names Claude Code, Codex, and LangChain agents.
- Agents can be configured by role, instructions, tools, and room membership, which limits their room context.
- Console telemetry can be disabled from the first-run notice or settings; server telemetry is off by default.
- Switch does not supply agents or models, so teams must run and connect their own choices.
- Remote server hosting is linked from the README, but the supplied material does not give deployment commands, complete runtime dependencies, or credential requirements.
- Anonymous Console analytics are enabled by default; server telemetry is controlled separately by TELEMETRY_ENABLED=true.
- The supplied material does not state software, model usage, or third-party tool costs.
How does this agent compare with similar options?
The README contrasts Switch with single-question AI bots and tools that aim to replace a team's workplace: Switch brings agents into existing chat apps and does not replace the chat app, agent, or model.
Key facts side by side with the most closely related agents.
| Agent | Source review | Form / cost | Stars | Updated | Language | Full support on |
|---|---|---|---|---|---|---|
| Switch AI Team Workspace This agent | 41 · Major gaps | Desktop app | ★ 3.8k | today | Python | — |
| Commonly | 44 · Major gaps | Self-hosted serviceFree + model costs | ★ 1.4k | today | JavaScript | Codex · Claude Code |
| OpenCrew Team OS | 48 · Major gaps | CLIFree + model costs | ★ 496 | 7mo ago | Shell | — |
| somora | 69 · Some gaps | CLIFree | ★ 35 | 9d ago | TypeScript | ChatGPT · Codex · Claude Code |
How does FollowAgents rate this agent?
Why each dimension lost points
README claims room/role-based control over which agent a user can talk to, but this is marketing-level only; no permission model, default permissions, or least-privilege configuration is evidenced in code, so least_privilege is 1. For user_confirmation, README describes agents auto-opening channels, filing tickets, and deploying to test environments with no confirmation/approval gating described, so 1. Data-flow transparency is stronger: README and docs/TELEMETRY.md enumerate telemetry fields, note Console telemetry is opt-out and Core is opt-in/off by default, and state sending stops immediately on opt-out, so 2. Sensitive data handling: only asserts no keys/prompts collected, no code-level evidence, and Console telemetry defaults on, so 1. Dependency security: only pinned action SHAs and pnpm frozen-lockfile in CI; no audit/SBOM/vuln-scan evidence, so 1. External effects: agents can deploy, edit runbooks, etc., with no rollback or sandboxing described, so 1. Rollback: only DB migration tests exist; no product-level rollback/undo mechanism, so 1. Source attribution: LICENSE carries copyright and Commons Clause notice, but no NOTICE or third-party attribution list in repo, so 1.
Self-consistency: README aligns broadly with CI and migration tests, but product behavior details (e.g., telemetry applying at restart) show tension and no end-to-end evidence, so 1. Dependency availability: only pnpm lockfile and Node version file references; actual lockfile contents or dependency manifests not shown, so availability cannot be confirmed, 1. Failure messages: CI has a version-format error message, but runtime error handling is unevidenced, so 1.
Audience and scenarios: README uses clear Level 1-4 layering across Slack/Teams/Discord/Telegram/Mattermost, so 2. Capability boundaries: a dedicated 'What Switch is not' section clearly states it is not a messaging app or agent provider, so 2. Trigger precision: rules for how agents are triggered in channels (@mention, auto-response) are not described, so 1. Environment fit: multi-platform installers for macOS/Windows/Linux plus self-hosting docs link, so 2.
Information architecture: README is well-structured with badges, layered explanation, Docs, Telemetry, Contributing, so 2. Install notes: two-step install plus per-platform download table and remote-hosting doc link, so 2. Naming stability: repo shows emdash, CHOO-1426, workspaceProvider→locationProvider renames, indicating evolving naming, so 1. Examples and FAQ: demo videos and scenario descriptions but no FAQ or troubleshooting, so 1. Known limitations: only indirectly via 'What Switch is not'; no dedicated limitations section, so 1. License: full Apache 2.0 + Commons Clause, clearly non-OSI, so 2. Versioning/changelog: release badge and RELEASING.md reference but no in-repo CHANGELOG, so 1. Maintenance responsibility: SECURITY.md gives response SLAs, but publisher identity is unverified and maintenance ownership is unclear, so 1.
Output usability: output is in-chat agent collaboration; README has demos but no concrete output samples or format specs, so 1. Marginal value: versus plain chat bots, Switch offers multi-agent room collaboration and shared context; value proposition is clear but lacks quantified comparison, so 1. Cost-benefit: self-hosting and bring-your-own-model can save cost, but no cost model or ROI data (only a marketing link to Switch ROI), so 1.
Claim traceability: many README claims ('context is yours', 'guardrails') point to marketing pages rather than in-repo evidence, so 1. Cross-source corroboration: README, SECURITY.md, CI, and migration tests are partly consistent, but core product behavior lacks code corroboration, so 1. Fact/inference separation: README mixes marketing claims with facts without distinguishing verified from unverified, so 1.
- README's claimed permission controls, guardrails, and cost reporting point to marketing pages with no in-repo code evidence; verify independently before enterprise adoption.
- Console telemetry is opt-out (on by default) while Core telemetry is opt-in; data-sensitive organizations should confirm settings before first run.
- Agents can auto-open channels, file tickets, and deploy to test environments with no described confirmation gating or rollback mechanism.
- License is Apache 2.0 + Commons Clause, a non-OSI license prohibiting selling derivatives; legal review needed for commercial use.
- Publisher identity is unverified; maintenance responsibility and update path are unclear.
- Naming is still evolving in-repo (workspaceProvider→locationProvider, etc.), raising API/config stability concerns.
Recent releases
Latest entries from the project's GitHub Releases. Notes are quoted from upstream as written, not rewritten or tested by FollowAgents.
-
switch-agent-controller-v0.31.1-dev.307.g194f132 switch-agent-controller 0.31.1-dev.307.g194f132 (dev build of main) Pre-release
Release notes excerpt
The Switch agent controller built from main at 194f1322146066ed2010f8ea4df876126de54b4f, for the dev environment, whose controllers update to it on their own. Not for installing by hand; switch-dev-cleanup prunes it. SHA-256: 58124009fc8ea65e3a501812cbea324b03527249d197542c72bfee65be58c45eView on GitHub ↗