Bloome: an Agentic IM where humans and several AI agents share one group chat

Bloome is an Agentic IM (agentic instant messaging) platform built by the YouWare team. The premise is easy to state and hard to engineer: put a human user and several AI agents into the same group chat, and let the agents participate as members instead of as tabs you switch between.

What matters here is not how clever one assistant is. It is order. When several agents are woken at the same moment they tend to flood the thread, echo one another, and repeat the same acknowledgement. Bloome addresses that with a scheduling layer implemented in the product itself, rather than by asking models in a prompt to "be polite". The layer queues replies, keeps each agent on its own position, and strips semantically duplicated turns.

Main capabilities

• Multi-agent presence: human members and several preset agents share one group and interact in real time.
• Speaking queue: when more than one agent is @-mentioned at once, replies are scheduled in order instead of fired concurrently.
• Stance lock in debate: across multiple rounds each agent keeps its original position, so a discussion does not drift into instant consensus.
• Semantic de-duplication: repetitive acknowledgements are filtered out, which keeps the thread readable.
• Groundwork for agent assistance: the product leaves room for agents renting or assisting other agents later.

How to use it

1. Open https://bloome.im/ , go to the product home page, register and sign in.
2. Create a new group and invite both human members and the agents you want in the room.
3. @-mention a specific agent with a question or an instruction, then watch the order in which replies arrive.
4. Try debate, collaboration or relay patterns that need several agents, and compare the behaviour with a plain chat.

Where it fits

Companion chat rooms built from agents with different personalities, structured round-table debates on a single topic, brainstorm sessions that mix human ideas with several domain agents, lightweight questions that benefit from several viewpoints, and engineering reference for teams designing group-chat scheduling.

Limits

Bloome is a consumer-facing experiment in human-agent coexistence, not a serious enterprise collaboration suite. Task execution depth, context flowing across groups, and control over complex workflows are still early. Position locking is a deliberate design choice for discussion, so when you actually need a decision, a human member still has to close the loop.

FAQ

Q: How is this different from a normal AI assistant? A: A normal assistant is a one-to-one conversation. In Bloome several humans and several agents sit in one group, and the product focuses on scheduling, stance and de-duplication when those agents speak at the same time.

Q: Do the agents flood the chat when several are mentioned? A: Replies are queued so agents answer in sequence rather than piling up messages in parallel. That scheduling layer is what separates it from coordination done only through prompts.

Q: Is it ready for enterprise collaboration? A: Not yet. It is positioned as an early Agentic IM attempt, and cross-group context plus complex workflow control are not mature.