User guide
Choose a focused RouteKit workflow for setup, integration, operation, or recovery.
RouteKit user guide
Use this page as a map. Each linked guide has one job, its own prerequisites, and a clear result, so you can read only the path that matches what you are trying to do.
For mutable route qualification and billing details, always consult routes and billing.
Start here
Install RouteKit
Install the CLI, configure one working route, and verify the local gateway.
Understand the architecture
Learn how the CLI, daemon, data plane, and control plane fit together.
If RouteKit is already running, choose the client or workflow you need below.
Connect a client
Connect coding tools
Launch qualified Codex or Claude Code builds, or install persistent native-client settings.
Check client compatibility
See the exact Codex and Claude Code builds qualified with RouteKit.
Call the HTTP gateway
Authenticate and send OpenAI, Responses, or Anthropic-compatible requests.
RouteKit's TypeScript packages are not tested or supported as a public embedded API. Applications should use the CLI-managed daemon and authenticated HTTP gateway. See TypeScript package status for the boundary.
Configure a route
Configure providers and models
Set provider credentials, the default model, aliases, catalog policy, and retention.
Pool subscription accounts
Enroll Claude Code or Codex accounts and configure same-provider rotation.
Set up AWS Bedrock
Apply least privilege, run preflight checks, and verify a Bedrock route.
Canonical model IDs are always provider/native-model. A request never falls
through to another provider or billing class when its selected route is
unavailable.
Operate or recover
Observe and operate RouteKit
Inspect health, account activity, quota, calls, logs, and retained rollups.
Run a remote gateway
Provision through SSH and connect through an operator-managed HTTPS front door.
Troubleshoot a failure
Diagnose daemon startup, missing models, subscription routes, remotes, and configuration.
Product lifecycle
Most RouteKit work follows the same five-stage loop:
- Configure a provider or enroll a subscription account.
- Verify live discovery with
routekit providers statusandroutekit models list. - Connect a coding tool or authenticated HTTP client.
- Observe health, usage, and call attribution.
- Diagnose with
routekit doctorand daemon logs when a route stops behaving as expected.
Use the command reference when you already know the task and only need exact syntax.
Before shared or production use
- Read routes and billing for the selected route's credential owner, egress, limits, and failover behavior.
- Read privacy and stored data.
- Use an exact build listed in client compatibility.
- Keep the default loopback bind unless an authenticated network boundary is intentional.
- Use named data-plane tokens instead of sharing the private owner token.
Develop from source
Contributors can follow develop from source for the repository toolchain, verification commands, and package map.