yetone drops Cumora, a Slack-like workspace where AI agents are your coworkers
yetone just open-sourced Cumora, a cross-platform team chat where AI agents work as first-class teammates alongside humans. Instead of waiting to be poked, agents hold persistent memory, claim Kanban tasks, coordinate with each other, and send real emails using cloud pods or local models. Why it matters: It moves AI from reactive chatbots to proactive office colleagues.
Cumora is an open-source, cross-platform workspace that turns AI agents into genuine, first-class coworkers inside a Slack-like environment.
Most apps treat AI like a glorified search bar where you have to ping a bot to get a reaction. Cumora flips this dynamic by giving agents their own persistent identity, long-term memory, and the autonomy to act on their own schedule.
What it actually does
In Cumora, human teammates and AI agents share the exact same roster, group channels, DMs, calendar, and Kanban board.
Agents don't wait around all day. They will proactively pick up tasks from the board, drop updates into group chats, start private discussions among themselves to make decisions, and even send and receive real external emails.
Outbound emails run through Resend, while inbound traffic passes through Cloudflare Workers, giving each agent a real communication channel beyond your chat room.
How it prevents AI chaos
Putting multiple self-directed agents in the same channel usually invites pure mayhem. Cumora solves this with a three-layer defense system.
First comes a seen-cursor freshness gate. If an agent takes too long to generate a response and humans drop new messages in the meantime, its draft is held and it is forced to re-evaluate the full thread.
Second is atomic task claiming. When a task hits the Kanban board, an agent must secure a distributed lock before touching it, ensuring two bots never duplicate work.
Third is small-brain triage. Routine mentions and background noise are handled by a lightweight model first. Cumora only wakes up expensive reasoning engines when an actual task demands heavy lifting, keeping your token costs sane.
Cloud brain or bring your own engine
Cumora ships with two distinct deployment paths.
The Cloud mode provisions dedicated Kubernetes pods for each agent, looping multi-hop tool calls directly through OpenAI APIs for bash commands, web browsing, and file edits.
The real highlight for technical teams is BYOA, or Bring Your Own Agent. You can link your local Mac or VPS with a single CLI command, wiring Cumora directly to Claude Code, Codex, Cursor Agent, or Gemini CLI. Claude Code and Codex stay inside strict sandbox boundaries, and your private API credentials never touch the remote server.
The stack is built on React 18, Node, PostgreSQL, and Redis. Running it locally only requires setting up Postgres, Redis, and an OpenAI key. The starter seed even populates six ready-to-run agents and test channels right out of the box.
Why it matters
We see Cumora as a clear signal that the era of passive chatbots is coming to an end. Autonomous peers in shared channels are the natural next step.
If you are an indie builder or running a lean software outfit, this is one of the cleanest open-source setups to bridge CLI dev agents like Claude Code with an everyday project tracker.
That said, giving AI the keys to your inbox and task board requires guardrails. Even with smart triage, autonomous chatter can burn through API credits quickly if you leave multiple agents unchecked overnight.
Sources (1)
- github.com(opens in a new tab)Originalgithub.com