Skip to content

About

Bitrouter documentations

Resources

Contributing

Stars

1 star

Watchers

0 watching

Forks

Latest commit

 

History

342 Commits

Folders and files

NameName
Last commit message
Last commit date
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

bitrouter-docs

The public BitRouter website — marketing pages, Fumadocs documentation, and the API reference — served at bitrouter.ai.

Develop

pnpm install
pnpm dev

Build

pnpm build   # prebuild regenerates the API reference from openapi.yaml + the models snapshot

Release on Railway

After CI passes on main, the release workflow deploys that commit to Railway staging and checks the homepage, quickstart, API reference, pricing, and changelog. Test the staging site, then open a release PR with production as the base and main as the compare branch. Merging that PR approves production.

The required Production release ready check verifies that the PR's exact source commit passed main CI and staging smoke checks. It also verifies that the proposed merge has the same source tree. If main advances while the PR is open, its head and checks update. Retest that new staging version before merging. If the check runs before staging finishes, rerun it after staging succeeds.

Use Create a merge commit to preserve shared history between release PRs. After the merge, Production release deploys the approved source SHA and runs the same smoke checks. Production runs are serialized, and a retry of an older release fails after a newer production PR is merged.

Configure the release workflow

  1. Create staging in the existing Railway project and connect its service to this repository. Assign it a separate HTTPS domain and set NEXT_PUBLIC_WEB_URL to that origin. Use staging credentials for any enabled integrations. The app sends X-Robots-Tag: noindex in this environment.
  2. Configure both services with the Railpack build, start command, and health check in railway.toml. New services use Railway service settings for these values. Railway ends legacy railway.toml support on December 1, 2026.
  3. In GitHub Settings → Environments, restrict staging deployments to main and production deployments to production. The release PR merge is the approval for production. In each environment set the RAILWAY_DEPLOY_TOKEN secret to a Railway Project Token for that environment. Add RAILWAY_DEPLOY_ENVIRONMENT_ID, RAILWAY_DEPLOY_SERVICE_ID, and RAILWAY_SITE_URL as environment variables. The production URL is https://bitrouter.ai.
  4. Disable Railway GitHub autodeploy for both environments and set the GitHub repository variable RAILWAY_STAGING_ENABLED to true.
  5. Initialize production at the currently deployed production source commit. Protect it with required PRs, Build docs site and Production release ready checks, administrator enforcement, and disabled force pushes and deletion. Merging a main → production PR starts the production workflow.

If a deployment times out, check its ID in Railway before retrying. If a production smoke check fails after Railway has switched traffic, use Rollback on the previous successful production deployment in Railway.

OpenAPI spec

openapi.yaml is the committed copy of the BitRouter API spec. It is the source of truth for the rendered API reference and is updated by an automated PR from bitrouter-cloud on each release — do not hand-edit it.

Public status page

/status reads one monitor-backed component from a public OpenStatus page with the server-side @openstatus/sdk-node client. Set OPENSTATUS_API_KEY to a read-only key, OPENSTATUS_PAGE_ID to the public page ID, and OPENSTATUS_COMPONENT_ID to the Inference API component ID. These variables remain server-only. Verify the component is linked to the intended monitor.

Without configuration or when data cannot be read, the page shows Status unknown. Old cached data is labeled out of date. Probe availability and synthetic response times describe OpenStatus checks; they are not an SLA or measurements of every inference request. Incident routes only expose reports listed in the configured public page/component snapshot.

Agent Playground

The homepage and /try share one fx conversation and virtual browser workspace. Visitors can draft before signing in to Cloud. Runs use real Cloud inference; there is no sample response or comparison mode. WebAssembly JSPI is required. Expand/back preserves the conversation, workspace and active task. Stop cancels execution. An unavailable service keeps the draft and reports an error.

To configure inference, set the server-only variables in .env.example:

  • BITROUTER_PLAYGROUND_ENABLED=1
  • A dedicated BITROUTER_PLAYGROUND_API_KEY and qualified BITROUTER_PLAYGROUND_MODEL; optional BITROUTER_PLAYGROUND_VISION_MODEL.
  • BITROUTER_PLAYGROUND_REDIS_REST_URL and BITROUTER_PLAYGROUND_REDIS_REST_TOKEN for a durable primary Redis REST endpoint. No extra service is provisioned.
  • An explicit BITROUTER_PLAYGROUND_QUOTA_ID, BITROUTER_PLAYGROUND_SITE_REQUEST_LIMIT, BITROUTER_PLAYGROUND_ACCOUNT_REQUEST_LIMIT and BITROUTER_PLAYGROUND_MAX_CONCURRENCY.

All instances must share the quota ID/store. Counts are lifetime admitted model requests for that campaign, including failures and cancellations. They do not reset when the browser resets, the service restarts, or a new session opens. Configure persistence and no eviction. One account can run one model request at a time; the site concurrency cap uses expiring leases. Missing configuration or a quota-store failure stops inference before the Cloud call.

These shared request/concurrency limits do not implement a USD spending ledger. Qualify the dedicated credential's funding/spending controls and the actual Cloud sign-in, tool-capable model, streaming and cancellation before enabling a hosted trial. See the playground contract. For a local operator run, use pnpm dev --hostname 127.0.0.1, the same quota configuration, and BITROUTER_PLAYGROUND_LOCAL_LIVE=1. That exception is disabled in production mode. The former BITROUTER_PLAYGROUND_CLOUD_PREVIEW flag is retired.

Email setup

The Enterprise contact form uses the server-only RESEND_API_KEY and RESEND_ENTERPRISE_FROM variables in .env.example. Configure a sending API key and verified sender on the Railway site service. Inquiries are delivered to contact@bitrouter.ai, with the visitor's address as reply-to.

The independent changelog announcer drafts email for highlight releases only. Configure its GitHub Actions RESEND_SEGMENT_ID variable for existing opted-in subscribers and RESEND_FROM for a verified sender; it uses the RESEND_API_KEY secret. Broadcasts stay as drafts until a human reviews and sends them in Resend. The website no longer provides a public email signup flow.

config reference

pnpm generate:config renders the local configuration reference and downloadable JSON Schema from a committed upstream snapshot. It runs at prebuild without network access or an upstream checkout. Edit config-reference/index.md for the page introduction and source-checked loading/reload guidance.

To update the schema, capture an explicit commit or tag from a BitRouter checkout:

pnpm snapshot:config --repo /path/to/bitrouter --ref <commit-or-tag>
pnpm generate:config
pnpm lint:config

Review the pinned revision and its runtime behavior before updating the prose. The schema does not prove released-binary support or encode reloadability.

About

Bitrouter documentations

Resources

Contributing

Stars

1 star

Watchers

0 watching

Forks

Contributors

Languages