Skip to content
CodeBook

Supported features and operating limits

What you can use in the hosted app, which setup each feature needs, and where CodeBook hands responsibility back to your team.

Choose the right setup

I want to… Start here What to check before relying on it
Run work from my computer Install the Claude Code plugin Run here depends on the local client and checkout. A signed-in chat connector by itself does not install a local worker.
Start work from a ticket, mailbox, alert or team chat Set up a factory plugin Match the workflow, provider account, allowed sender or issue label. Test the connection, then send a real example in your test project. A credential test does not prove delivery.
Catch the recurrence of a production bug Incident checks Replay the broken and fixed commits, then have the owner approve the rule. Make the trusted CI job required at your Git host to cover PRs outside CodeBook.
Send a reviewer the evidence for a GitHub change Pull requests Drafts show checks at the opening commit. The host's Checks tab shows later runs. Review minutes are reported by reviewers; a missing report is not zero effort.
Run a supervised job after closing chat Cloud work orders Agree the goal, deadline, budget, acceptance checks and decisions that still need a person. The spend limit covers supported execution at configured rates; your own CI and deployment providers are billed separately.
Watch a deployment for customer impact Production monitoring Connect the source and deployment, then enable the watch on the order. A pause requests an owner's decision. Your deployment system performs any rollback.

Assistant and provider coverage

Claude Code provides the local plugin. Claude chat and ChatGPT use the personal connector. Its permissions follow the signed-in person. The native plugin engine passed its regression suite on Windows and macOS. Windows also passed a clean installation and ran /factory help. With a local test connector, the installed Windows CLI listed a workspace’s factories and opened a specific workflow backlog page. Sandboxed local proof uses macOS or Linux, including WSL with bwrap. An uninterrupted 16-hour customer run and a paid customer pilot have not been verified.

Test third-party integrations with the accounts and permissions you intend to use. Local integration tests cover ticket and chat workflows, but do not certify delivery in a customer's Jira, Linear, Teams, Telegram, WhatsApp or email account. Worker SDKs for JavaScript/TypeScript and Python are available from source; follow the source installation instructions for the client you use.

Factory versions and evidence

A factory export or rollback restores the supported definition. It is not an environment backup. Pinned knowledge revisions, all historical skill dependencies, tool grants, runtime bindings and worker-specific model policy are not all recreated by importing a ZIP. Keep external credentials and runner configuration separately. A canary comparison is evidence about the measured runs; validate promotion on your deployment.

Use proof runs to choose the runner and checks that apply to a repository. A passing test covers the measured behavior and revision. A missing result, an unpriced operation or a provider that cannot enforce the work order's limits is a reason to inspect the order, not evidence for approval.

Who operates what

Bookbag manages identity, membership, provider credentials and billing. Workspace permissions, station approvals and factory controls still apply in CodeBook. See organization controls for audit, retention and telemetry. CodeBook requires a reachable Bookbag SSO; an offline or air-gapped installation is not supported. Teams operating a copy in their own cloud can use the deployment runbook for PostgreSQL, SSO and startup checks.