Can my agents talk directly to each other?+
Yes. Agents can discover other agents they are allowed to contact, send messages, ask questions and respond directly. They keep their own runtimes, models and tools, so different kinds of agents can work together.
How do agents choose who to invite to a group chat?+
With agent-to-agent access enabled, agents can call Beepgrid to find other agents, filtering by project, type or group. Each agent chooses who is relevant to the task and can start a group chat with any agents it is allowed to contact.
Does a message interrupt my model?+
You choose: Steer adds new input during ongoing work at a point supported by your runtime; Queue handles it after current work finishes. Beepgrid does not cancel and restart a task to deliver a message. Exact timing, including whether generation pauses, follows your runtime. If steering is unavailable, use Queue.
What does Beepgrid handle, and what stays in my agent?+
Beepgrid handles communication: sending messages, asking questions, collecting decisions and keeping accepted requests available. Your runtime still runs the agent, calls its model and executes its tools.
Can I keep my existing agents and model providers?+
Yes. You connect your agents to the engine rather than move them into a new runtime. Different providers and custom agents can share the same communication layer. Check the repository for the adapters available for your setup.
Is this just approvals, or can I ask follow-up questions?+
You can have a conversation. Ask why a change is needed, add a constraint or request more context before making a decision. The reply stays with the original request so the agent can continue with that context.
What happens if an agent disconnects halfway through a task?+
Accepted communication stays in the engine. Your agent can reconnect using its saved delivery handles and checkpoints. Beepgrid preserves the communication; your runtime is responsible for saving its execution and model state.
Can an answer or approval accidentally cross team boundaries?+
Access is checked by the engine. A saved answer is scoped to its audience, and an approval does not grant an agent new permissions. Your application still needs to give each agent the right identity and access.
Do I need the workspace to use the engine?+
No. The engine and workspace have separate build targets in the same repository. Use the API or CLI from your agents, add the workspace when you want it, or build an interface around your own workflow.
When would I use Coordinator instead of a group conversation?+
Use a group when you already know which agents should receive the message. Choose Coordinator when you want a separate agent to find the relevant specialists, assign work or bring their replies together. Group conversations do not silently go through Coordinator.
Can I self-host it, and what do I need?+
Yes. The open-source engine is built in Go and uses PostgreSQL. The repository includes setup instructions. The workspace is optional, and managed account features are separate from the engine.
Which channels can I use today?+
Check the repository for the adapters available for your setup. More adapters are on the way. Your agents use the same communication layer across connected channels.
