Skip to main content
Goal: Migrate an existing app’s LLM calls to OpenRouter without rewriting how the app works. Outcome: Your model calls route through OpenRouter with the same streaming, tool calling, and response shapes, using your existing framework wherever a clean OpenRouter path exists. If your app has server-side agent or tool loops, the prompt can optionally move that execution onto OpenRouter’s Agent SDK (@openrouter/agent). That’s a bigger change than a provider swap, so the prompt recommends it only when it fits and asks before doing it — you can decline and keep the minimal migration. OpenRouter is OpenAI-compatible, so most migrations come down to three changes: the base URL, the API key, and the model ID. The prompt below does that work for you across whatever stack your app uses. Paste it into a coding agent at the root of the project you want to migrate. The prompt comes in two versions. Use the With /goal tab if your coding agent supports a /goal command — it leads with an explicit goal and done-criteria so the agent runs the migration to completion instead of stopping at advice. Use the Without /goal tab to paste the migration prompt on its own.

​
Prerequisites

  • A coding agent with file access to your project (Claude Code, Cursor, Codex CLI, OpenClaw, or similar)
  • An OpenRouter API key — create one at openrouter.ai/settings/keys, and add credits if your models aren’t free
  • A clean git working tree so you can review the diff before committing
This prompt edits your codebase. It changes provider imports, model IDs, environment variables, dependencies, and docs, then runs your lint/typecheck/test/build commands. Run it on a branch and review the diff before merging. If you let it run a live smoke test, that request spends OpenRouter credits.

​
How to use it

  1. Open your coding agent at the root of the project you want to migrate.
  2. Make sure OPENROUTER_API_KEY is set in your environment. The prompt checks for it without printing the value.
  3. Paste the prompt and let the agent inventory your AI integration code first.
  4. If your project runs server-side agent or tool loops, the agent may ask whether to include an Agent SDK migration. Say yes to move that execution to @openrouter/agent, or no to keep the minimal provider swap.
  5. Review the migration plan and the diff. The agent preserves your framework and retargets only the model-call boundary.
  6. Run your tests and hit a real route to confirm behavior is unchanged.

​
What the prompt handles

The prompt picks a different path per stack so the migration stays minimal:
  • OpenAI SDK (TypeScript or Python) — retarget baseURL / base_url to https://openrouter.ai/api/v1 and swap model IDs. See OpenAI SDK.
  • Vercel AI SDK — swap the provider for @openrouter/ai-sdk-provider while keeping generateText, streamText, tools, and UI stream protocols intact.
  • Vercel AI Gateway — replace gateway(...) and AI_GATEWAY_API_KEY with the OpenRouter provider and OPENROUTER_API_KEY, including the user-facing setup text.
  • Mastra — keep Mastra and switch model strings to openrouter/provider/model. See Mastra.
  • Anthropic SDK — translate Claude message content to OpenRouter chat completions or the native OpenRouter SDK.
  • Anthropic Agent SDK — keep the agent loop and point it at OpenRouter with ANTHROPIC_BASE_URL and ANTHROPIC_AUTH_TOKEN. See Anthropic Agent SDK.
  • Raw HTTP and unlisted stacks — retarget OpenAI-compatible calls and preserve route shapes, streaming, and tool semantics.
OpenRouter Agent SDK (optional) — when your project has server-side tool execution, multi-turn agent loops, or custom orchestration, the prompt can move that boundary to @openrouter/agent. It recommends this only when the evidence is there and asks before making the change; otherwise it stays on the least-disruptive provider path above. For other new OpenRouter code, the prompt prefers @openrouter/sdk for platform APIs and the openrouter package for Python.

​
Check your work

  • Your lint, typecheck, test, and build commands pass on the migrated code.
  • Model calls hit https://openrouter.ai/api/v1 using OPENROUTER_API_KEY, with OpenRouter provider/model slugs.
  • Streaming, tool calls, and structured outputs behave the same as before.
  • No stale provider keys, base URLs, or model IDs remain, except boundaries the agent deliberately left in place and documented (such as embeddings or image generation).

​
Next steps