yetone Open Sources Cumora: AI Agent Becomes a Formal Member of Chat
Developer yetone has open-sourced the Cumora project, positioning the AI Agent as a formal member of team chats.
The interface is similar to Slack, with the Agent having a name, persona, memory, the ability to chat privately, create groups, claim tasks, send and receive real emails, and initiate conversations proactively.
It supports two operating modes: Cumora Cloud uses the OpenAI Responses API in a cloud Kubernetes Pod; BYOA allows the Agent to use the user's own Claude Code or Codex CLI through local commands.
The coordination mechanism includes outdated response interception, atomic task claiming, and lightweight model triage to avoid conflicts between multiple Agents and ineffective token consumption.
Currently in invitation-only beta testing, it fully supports local deployment, with desktop coverage for macOS, Windows, and Linux.
In terms of market dynamics, competition among event-driven multi-Agent collaboration tools is intensifying: funding and developer attention are shifting towards persistent Agent workspace solutions, putting traditional chat tools and single-point AI assistants under pressure in human-machine hybrid scenarios.
Additionally, it provides default roles such as Atlas and Bram, and supports boards and calendars.
Source: Public Information
ABAB AI Insight
Cumora transforms AI from a passive dialogue box into a team member with a persistent identity and proactive behavior, reconstructing the human-machine collaboration interface.
In terms of capital pathways, resource investment is directed towards both cloud Pods and local engines, with a coordination layer aimed at reducing multi-Agent conflict costs and supporting self-hosted privacy needs.
Similar cases can be seen in Slack's integration of AI assistants and various Agent framework collaboration experiments; it is currently evolving from single-Agent tools to multi-Agent persistent workspaces.
The structural judgment indicates a technological replacement: the chat interface shifts from a human-centric model to one where human and machine are equal members, with mechanisms of memory, proactive triggers, and conflict arbitration enabling Agents to have sustained working capabilities rather than one-time responses.
ABAB News · Cognitive Laws
Persistent identity is more important than one-time answers
Coordination mechanisms determine multi-Agent availability
Interface forms define collaboration boundaries