@powerhousedao/ph-clint-openrouter

A ph-clint plugin that adds OpenRouter and any OpenAI-compatible provider as an opt-in alongside the Anthropic default. It exposes resolveModelConfig, a pure function that maps a ph-clint runtime config (model, modelBaseURL, modelApiKey, per-provider keys) into a Mastra MastraModelConfig.

It is an external counterpart to default ph-clint codegen: core and the codegen builders stay Anthropic-only, so multi-provider support stays out of them. A consumer that wants it pulls in this package and calls resolveModelConfig from its own createAgent factory — no provider logic leaks into core.

Install

pnpm add @powerhousedao/ph-clint-openrouter

Peer dependencies: @powerhousedao/ph-clint and @mastra/core.

Usage

import { Agent } from '@mastra/core/agent';
import { resolveModelConfig } from '@powerhousedao/ph-clint-openrouter';

const agent = new Agent({
  id: 'my-agent',
  name: 'My Agent',
  instructions: '…',
  model: resolveModelConfig({
    model: ctx.config.model, // e.g. 'openrouter/openai/gpt-4o'
    modelBaseURL: ctx.config.modelBaseURL,
    modelApiKey: ctx.config.modelApiKey,
    anthropicApiKey: ctx.config.anthropicApiKey,
    openrouterApiKey: ctx.config.openrouterApiKey,
  }),
  // …
});

Resolution

resolveModelConfig(cfg) returns either an OpenAI-compatible { id, url?, apiKey? } object or a bare model string. It is structurally assignable to Mastra's MastraModelConfig.

| Input | Output | | ------------------------------------------------------ | --------------------------------------------------- | | anthropic/* + key | { id, apiKey } | | anthropic/*, no key | bare model string | | openrouter/* + key | { id, url: openrouter/api/v1, apiKey } | | openrouter/*, no key | bare gateway string (Mastra reads OPENROUTER_API_KEY) | | any modelBaseURL set | { id, url, apiKey? } | | otherwise | { id, apiKey } with a key, else bare model string |

Key precedence: a generic modelApiKey overrides the legacy <provider>ApiKey field (e.g. anthropicApiKey, openrouterApiKey).

OpenRouter model ids

Mastra strips the leading openrouter/ provider segment before calling the gateway, so the remainder must itself be a valid OpenRouter slug (<vendor>/<model>) — e.g. openrouter/openai/gpt-4o is sent as openai/gpt-4o. OpenRouter's own namespaced models (the routers openrouter/free / openrouter/auto, and stealth models) are themselves slugs, so write the prefix twice: openrouter/openrouter/free. A bare openrouter/free would be sent as free and rejected with an opaque 502, so resolveModelConfig throws a descriptive error for that shape instead.

License

MIT