Short definitions of the terms you’ll see across these docs, with links to the full explanations.
| Term | Meaning |
|---|---|
| Automation | An autonomous loop bound to a repository and a schedule that spawns implementations without anyone present. See Automations. |
| Blueprint | The workflow template a run follows. Task blueprints include bug fix, feature, and quick change. |
| Brief | A reviewed specification an agent works from (draft → approved → archived). See Briefs. |
| Hook | A command run at a sandbox lifecycle point. The SandboxReady hook runs setup after the sandbox is ready. See Hooks. |
| Implementation | One autonomous run that takes a task from plan to pull request. See Tasks & implementations. |
| Insights | Workspace analytics on pull-request activity, Trevize’s contribution, and cost. See Insights. |
| Integration | A connected external service (GitHub, Linear, Claude Code, MCP, Neon, Sentry, Supabase). See Integrations. |
| Knowledge | Durable observations agents capture as they work, searchable by later runs. See Knowledge & memory. |
| MCP | Model Context Protocol — the Trevize MCP server connects Claude Code to your workspace. See MCP. |
| Observation | A single unit of captured knowledge (decision, bug fix, feature, refactor, discovery, or change). |
| Preview | A live, running view of your app served from the sandbox. Configured with preview targets & processes. |
| Preview process | One command + port within a preview target (e.g. web, api). |
| Preview target | A named group of processes that together form a previewable app. |
| Pull request | The GitHub PR every run produces for you to review and merge. |
| Resource | A managed dependency (e.g. a database branch) provisioned per run and mapped into the sandbox. See Resources. |
| Run | A single execution of agent work — either a collaborative session or an autonomous implementation. |
| Sandbox | The isolated cloud environment where an agent checks out your repo and works. |
| Session | A live, collaborative run where humans and an agent build together in real time. See Sessions. |
| Task | A single unit of work that runs autonomously to a pull request; realized as an implementation. |
| Turn | One unit of agent work within a session; the agent runs one turn at a time while further instructions queue. |
| UI feedback | Comments pinned to a live preview that the agent can act on. See Reviewing implementations. |
| Workspace | Your team’s boundary — it owns repositories, members, and billing. See Workspaces. |