Definition
An agent profile is a named preset for the chats an AI agent asks to start, setting which app and model run it, how much effort it spends, what it may do, and when it's the right choice.
Also called: profile, helper profile.
When one agent delegates to another, someone has to pick the model and the permissions. Hard-coding model names into prompts goes stale, and letting the agent choose freely invites it to pick the most capable option every time. A profile gives the choice a name ("Deep reviewer", "Scout") that the person maintains and the agent only refers to.
Profiles are close to Claude Code agents and subagents, which also bundle a model and tools. The difference is scope: a profile describes a whole separate chat, possibly in another app, rather than a helper inside the same session.
In Pairbar
Settings → Integrations → Profiles starts with five: Deep reviewer (Claude, strongest model, high, read-only), Implementer (Codex, default model, medium, can edit), Second opinion (the other app, strongest model, high, reads files only), Scout (whichever app has more plan left, the small model, low, reads files only) and Writer (Claude, default model, medium, can edit). Agents see them in status and name one; Pairbar fills in the rest. What a request says itself wins, and a profile's access is capped at what the asking chat may do. Profiles are kept in ~/.config/pairbar/profiles.json. See set up agent profiles.
Frequently asked questions
- Can a profile give a helper more access than the chat asking?
- No. In Pairbar a profile's access is a default, capped at what the asking chat may do itself.
- Which profile does a review use?
- Deep reviewer, unless the chat names another. Second opinions use Second opinion. Both still go to the other app than the asking chat's.
- Is a profile the same as a Claude Code agent?
- No, but a profile can name one: its instructions then come first in the helper's chat.
Related
- GuideSet up agent profiles for second opinions, reviews and helper chatsPresets for the chats agents ask for (app, model, effort, access, when to use it): the five defaults, how Pairbar fills in a request, edit and reset.
- DocsIntegrationsLive status for Claude and Codex through their hooks, Pairbar's tools for their chats (the plugin, the `pairbar` command, the tool reference, who's calling, agent activity), and agent actions: second opinions, reviews, helper chats, handoffs, workflow runs and drafts, and check-ins that agents ask for and you approve, with budgets and guardrails
- GlossaryClaude Code agentA Markdown file with YAML frontmatter that defines a reusable Claude Code specialist, such as a reviewer, Claude can delegate to.
- Use caseHave Codex review Claude Code's changes, and Claude review Codex'sA chat asks for a read-only review of its uncommitted work, branch or PR by the other app; Pairbar reads the diff itself and returns the verdict.