An open-source (MIT) experimental platform by developer yetone that blends team chat with multi-agent collaboration. The interface resembles Slack or Discord, but the “members” include AI agents with names, personas, and persistent memory. These agents can claim tasks, coordinate with each other to avoid collisions, send real emails, and proactively initiate conversations based on schedules or messages. Two execution modes: Cumora Cloud runs agents in hosted Kubernetes pods via the OpenAI Responses API; BYOA (Bring Your Own Agent) runs a daemon on your own Mac or VPS, allowing agents to use your local Claude Code, Codex, or Cursor Agent subscriptions and API keys (the server never sees your keys, and agents run in a sandbox by default). Critical caveats: This is a very early-stage project—npm weekly downloads are near zero, the iOS app is in TestFlight, Android is unreleased, and there is no public pricing model. BYOA requires Node.js, local environment setup, and your own LLM subscriptions, making it a developer playground rather than a ready-to-use productivity tool. Independent testing (UCloud) found it suitable for first-round discussion and task breakdown, but not for high-risk decisions or sensitive data.

Pros & Cons:

Pros: The “chat as a multi-agent workspace” concept is genuinely novel—agents have persistent identities and can coordinate within a familiar chat interface, which is a different interaction model from CrewAI’s code-first orchestration or n8n’s node-based workflows; the BYOA architecture is a clever privacy and cost play—you can route agents through your existing Claude Code or Codex subscriptions, and the server never touches your API keys; it is fully open source (MIT), so the collaboration logic and agent harness are auditable and forkable; the project is backed by a known developer (yetone, creator of avante.nvim) with a track record of building AI developer tools.

Cons: It is extremely early-stage and unstable—npm weekly downloads are effectively zero, mobile apps are in beta or unreleased, and long-term maintenance depends on a small team or individual, making it unsuitable for production or daily reliance; setup friction is high—BYOA requires Node.js, a local daemon, and your own Claude Code/Codex subscriptions, while Cloud mode requires an OpenAI API key and involves data leaving your machine; there is no public pricing—it is unclear how Cloud usage is billed (your own API key vs. platform-managed), and no commercial support exists; independent validation is thin—only a light UCloud review exists, which explicitly warns against sensitive data or high-stakes use; the target audience is narrow—it is a developer experiment, not a tool for non-technical users seeking a ready-made AI assistant.

Who is it for?:Developers and AI tinkerers who are comfortable running local daemons, managing their own API keys, and experimenting with multi-agent collaboration; users who already pay for Claude Code or Codex and want a shared chat interface where agents can operate on their behalf; open-source enthusiasts who want to inspect, fork, or contribute to a multi-agent workspace concept. Not suited for non-technical users seeking a ready-to-use productivity app, teams requiring stable uptime or enterprise support, anyone uncomfortable with local environment setup, or organizations handling sensitive data (independent testing explicitly advises against it).