Skip to content
Critique/docs
Platform

Chat & Workspace

Use Critique Chat for repo-grounded Q&A, workspace exploration, and optional builder jobs.

Beyond automated PR review, Critique offers interactive surfaces for questions, exploration, and follow-up work on connected repositories.

Critique Chat (/chat)

Chat is a conversational interface grounded in your selected GitHub repository (and optional review-run context). Use it when you need to:

  • Trace how something is implemented without spelunking the tree manually.
  • Summarize a PR, CI failure, or recent review run.
  • Draft issues or PR comments (with confirmation before anything is posted).
  • Start Remedy after you have explicitly decided to fix findings.

The assistant chooses tools automatically — repository search, PR inspection, memory search, docs lookup, and others. See Chat tools for a full list and when each applies.

Chat models are configured under Settings and are separate from the models used in automated review policy.

Workspace (/workspace)

Workspace is Critique’s unified signed-in surface — one left rail for Chat and Builder history, one repo context bundle, and one credit ledger across review handoffs, Remedy, and build mode. Open /workspace directly, or escalate from Chat when a thread needs more room or a structured handoff into build mode (?mode=build).

For the product story and what shipped in May 2026, see the launch essay: Introducing Critique Workspace.

Think of chat as the conversation and workspace as the working surface around it.

Agent queue

Stage prompts before starting a long run:

  • Modes: chat (Ask) or build (Builder-style execution)
  • Runtimes: Critique agent, Claude Code, or Codex (labels in the UI)
  • API: GET / POST / DELETE on /api/workspace/queue and /api/workspace/queue/{queueId} (session auth)

Items are scoped to your active repository and optional chat or builder session.

Inspector → Processes

The workspace Processes panel shows live state for streaming chat, builder jobs, sandboxes, and retrieval — so you can tell whether work is still moving without opening raw logs first.

Builder (/builder)

Builder runs cloud jobs in isolated OpenCode + E2B environments — useful for larger changes that should not block the chat UI. Jobs appear under Dashboard → Builder (and in the workspace rail) with status, activity events, and diffs.

Builder and Remedy share a similar “execute in the cloud” idea, but Remedy is tied to a specific review run and bounded fix scope; Builder is for general agentic tasks you initiate from the product.

Coding Agent API

The same runtime is available over HTTP for automation: Coding Agent API (POST /api/v1/coding-agent/runs, follow-up messages, optional draft PR). Requires crt_ keys with read:builder / write:builder scopes.

Credits and availability

Chat and builder workloads consume credits like reviews. Lightweight Q&A is typically cheaper than a full PR pipeline; starting Remedy or builder jobs costs more. Check Dashboard → Usage if spend surprises you.

Some tools (external doc search, handoff to Claude Code) require optional integrations or API keys in Settings.

Privacy note

You can disable chat history retention in account-related settings when your organization requires it. Repository indexing for search and memory follows installation scope — only repos you granted the GitHub App can be read.