The chat assistant lets you work with Coroid conversationally rather than through forms. Find it under Chat.
What it is good at
Starting work from a half-formed idea. When you know the outcome but not how to phrase it as a specification, a conversation gets there faster than a blank field — you can be asked clarifying questions instead of guessing what to include.
Asking about state. What is running, what is blocked, why did this fail. The answer arrives without navigating to find it.
Asking about the codebase. Where something is handled, what depends on what.
Steering work in flight. Adding context to a running task, or answering a question it has raised.
It uses the same governed tools
The assistant is not a separate mechanism with its own privileges. It works through the same tool registry as everything else, so anything it starts goes through the normal path — specification, plan, verification, review, pull request.
This matters: asking conversationally does not bypass approvals. A task started from chat still produces a specification you approve.
Conversation is not specification
The one habit worth forming: when the assistant produces a specification from your conversation, read it as a specification.
A conversation is forgiving — you can clarify as you go. A specification is executed literally. Things that felt settled in conversation sometimes land vaguely in the written version, and it is the written version that gets built.
See Specifications.
Chat in your own tools
Where a chat platform is bridged, some of this is available in the channel your team already uses. See Chat platforms.
For programmatic access from your own assistants and editors, Coroid can act as an MCP server — see Coroid as an MCP server.