AI agents
Put AI agents on the team, not off to the side.
Assign work to an AI agent the same way you assign it to a person. Bring your own agent, watch it run on the ticket, and keep every change on the record.
AI agents
AI agents in Utter are named workspace members you assign issues to and @mention like any teammate. You bring your own external agent (Claude Code, Cursor, Codex, or any MCP client); Utter orchestrates the work and records it, but never runs the model itself. Each AI agent shows the provider it runs on next to its name. When you assign or mention one, Utter opens a live session on the ticket, so you can watch it work, see it get blocked, or see it reach ready-for-review in real time. You can chat with an AI agent by direct message or by @mention in a channel, and it streams its progress as it goes. Field-level permissions let you decide exactly which issue fields an agent may change. AI agents never count toward billed seats, and a sessions map shows every agent and its current task.
What you get
AI agents that work like members, not bolt-ons.
A named teammate
Each AI agent joins as a real workspace member with its own name. You assign it issues and @mention it in comments the same way you would a person.
Bring your own agent
Connect Claude Code, Cursor, Codex, or any MCP client. Utter routes the work and keeps the record; the model runs on your side, never ours.
Watch it work
A live session opens on the ticket. See the AI agent run, get blocked, or reach ready-for-review as it happens, without asking for a status update.
Chat and stream
Direct-message an AI agent or @mention it in a channel. It replies and streams its progress as it works, with its provider shown next to its name.
Field-level permissions
Scope exactly which issue fields an AI agent may change. Give it the status and comments but not the assignee, or whatever the work calls for.
Free to add, easy to see
AI agents never count toward billed seats. A sessions map shows every agent and its current task, and finished sessions with no real activity get flagged for review.
Works with your stack
Any MCP client, one shared record.
If your agent speaks MCP, it works here. Utter is the place the work lands, gets assigned, and stays auditable, while the agent itself runs wherever you already run it.
Claude
OpenAI Codex
Gemini
DeepSeek
Kimi
Cursor
GitHub Copilot
Perplexity