Your context graph, mounted
Memories, runbooks, and conventions load into every session. The agent codes the way your team codes.
Recorded with the runLoaded source list
Surfaces
Context Code & CLI
The Context agent, terminal-first.
Run Context from the command line: a coding agent with your context graph mounted, usable interactively, in scripts, or headless in CI. It arrives already knowing your conventions, runbooks, and past corrections.
Context Code & CLI
Example workflow
context / checkout-servicerunning
› Add idempotency to refund creation and cover it with tests
Reading repository instructions and relevant call sites…
Plan approved
1. Trace refund creation path
2. Add idempotency key storage
3. Cover retry and collision cases
✓ pnpm test refunds
8 passed · 0 failed · 4.2s
repo write · tests · no networkReview diff
Changed files
refunds.ts+24 −6
idempotency.ts+48
refunds.test.ts+61
Loaded context
API conventions
Payments runbook
3 prior corrections
How it works
From request to a reviewable result.
Each run keeps the request, actions, evidence, and final decision together.
01 · Scope
Give Context a repository task.
Start an interactive session or a headless run. Context loads the repository rules and approved team context.
Output · Plan and permission scope
02 · Execute
Watch the files and commands.
Changes run in a sandbox while the terminal reports tool calls, tests, and any decision that needs you.
Output · Traceable working diff
03 · Review
Keep the result you want.
Inspect the patch and test results, request another pass, or hand the change to your existing review process.
Output · Patch and run report
What you can inspect
Capability is only useful when it leaves evidence.
Headless CI mode
Deterministic runs with exit codes and machine-readable reports, for pipelines and pre-merge checks.
Recorded with the runExit code and JSON report
Sandboxed by default
Sessions execute in Agent Sandboxes with scoped credentials, not on shared runners with standing secrets.
Recorded with the runSandbox and credential scope
Traced like everything else
Every session produces a full trace, gradeable against your rubrics and usable as training signal.
Recorded with the runCommands, diff, and grade
Deployment boundaries
Control is part of the product.
Context Code & CLI can run as a standalone surface, while its traces, evals, and policy decisions remain part of the same Context system.
Review stays with your team
Context prepares and tests a change; your repository protections still decide what can merge or deploy.
Context is selected
Only approved repositories, memories, and connected systems are mounted into a session.