Multica / Collaborate with agents
dialogue
Chat one-on-one with the agent independently - completely sandboxed, the agent cannot see or change issues, and no one can see your conversations.
Overview
Chat is a one-on-one independent communication between you and the agent - after jumping out of the issue board, the agent cannot see any issue or change the issue. The entire conversation is completely private (other people in the workspace, including the admin, cannot see it). Suitable for discussing plans with agents, doing brainstorming, and asking a question that does not belong to any issue.
Why not use @ agent?
@mention pulls the agent into the context of the issue - it will read the issue's description and all historical comments, and can also modify the issue. The conversation is reversed: pulling you out of the issue - the agent can only see this conversation, and does not know that the issue exists, and has no entry to modify the issue.
Two criteria:
- Ask the agent to give feedback based on the context of a specific issue → @mentioning
- Want to chat with the agent about a topic that does not belong to any issue (or do not want anyone to see the discussion) → Dialogue
start a conversation
Enter the Chat portal from the sidebar, select an agent, and open a new conversation. The interface is the same as that of ordinary chat software: you send a message and the agent replies. Each message triggers an execution in the background (enqueuing a task), so the reply may take several seconds.
What an agent can and cannot do in a conversation
The agent runs in a completely sandboxed conversation.
What you can do:
- Answer questions asked in your current message
- Use your own configured skill and MCP
- Read and write files in your own working directory
- Call
multicaCLI commands that do not require issue context (such as querying basic information about the workspace)
What not to do:
- See any issue - there is no issue ID in the prompt received by the agent, and commands such as
multica issue listreturn empty to it - Change any issue - without issue context, API calls will be intercepted by permission check
- See other conversations - Complete isolation between conversations
- @any person or agent - the conversation is a private space, there is no way to notify others
How to retain context in multi-turn conversations
The conversation uses the provider session recovery mechanism to maintain multiple rounds of context - when the agent responds for the first time, a provider session (such as Claude's session) is established, and the session ID is saved; when the next message is dispatched, the ID is sent back, and the agent continues with the last state without re-reading the history each time.
If a certain round fails, Multica will look for the task that created the session ID in the previous round (whether it succeeded or failed at that time) and try to resume - the memory of the entire conversation will not be lost because of an error in the middle.
Note: Not all providers actually implement session recovery - see Providers Matrix for support.
Archive conversation
Conversations you no longer want to see can be archived - right-click on the conversation list or click the "Archive" button on the details page. After archiving:
- Conversations are hidden from the active list (can be accessed in "Archived" view)
- Historical messages, session IDs, and working directories are kept intact and will not be deleted.
No "Restore" button after archiving - There is currently no way to reset archived conversations back to the active portal. If you want to continue this conversation later, you can only start a new conversation. When you need to browse the contents of an archived conversation, go to the "Archived" view to read historical messages.
Next step
- Autopilots - Let the agent automatically start work at a scheduled time
- Assign issues to agents - Put the topic back on the issue board