Usage for Agents | JoyToken

Use this page when you want Codex, Claude Code, OpenClaw, Hermes, or another coding agent to modify an application that calls JoyToken through the Client SDKs.

What agents should know

Topic	Instruction
Package choice	Use `@joytoken/client-sdk-ts` for TypeScript projects and `github.com/joytoken/client-sdk-golang` for Go projects
API key	Read `JOY_TOKEN_API_KEY` from server-side environment variables
Model	Start with `auto`; use a concrete JoyToken model key only when the workflow needs a pinned model
Streaming	Treat chunks with `metadata` as observability data, not assistant text
Errors	Handle `JoyTokenAPIError` in TypeScript and `APIError` in Go
Browser code	Do not put JoyToken API keys in frontend bundles

TypeScript prompt snippet

agent-instructions.md

Use @joytoken/client-sdk-ts for JoyToken model calls.
Create a JoyTokenClient with JOY_TOKEN_API_KEY from server-side env.
Use chat.completions.create() for non-streaming calls.
Use chat.completions.stream() for SSE streaming.
Do not expose JOY_TOKEN_API_KEY in browser code.
Handle JoyTokenAPIError and log status, requestId, and body.

Go prompt snippet

agent-instructions.md

Use github.com/joytoken/client-sdk-golang for JoyToken model calls.
Create a client with joytoken.NewClient(joytoken.WithAPIKey(os.Getenv("JOY_TOKEN_API_KEY"))).
Use CreateChatCompletion for non-streaming calls.
Use StreamChatCompletion for SSE streaming and close the stream.
Handle *joytoken.APIError with errors.As.

Code review checklist

Check	Why it matters
Server-side API key	Prevents secret leakage
Request IDs	Makes gateway logs, usage, and app logs easier to correlate
Streaming metadata handling	Prevents metadata JSON from appearing in the UI
Error classification	Separates policy, quota, auth, and provider failures
Model choice	Keeps routing flexible unless a workflow requires pinning

For coding-agent product setup, use the Cookbook pages. This page is only about how agents should edit code that uses JoyToken Client SDKs.