@volter/twin-ai-gateway
v1.0.0
Published
Vercel AI Gateway twin (Protocol 3): its recorded model catalogue, OpenAI's chat completions and the AI SDK's language-model protocol on the World's scenario or the kernel wire's labeled stub (the twin runs no model), with API keys. Built on @volter/world
Readme
@volter/twin-ai-gateway
Vercel AI Gateway (Protocol 3), for LibreChat's configured Vercel custom endpoint and applications using the published AI SDK gateway protocol. Demand records the current application commits and exclusions. SOURCE records the reference-derived spec and catalogue capture.
It serves the recorded model catalogue at /v1/models, OpenAI-compatible Chat Completions at /v1/chat/completions, and AI SDK LanguageModelV4 generate results and SSE parts at /v4/ai/language-model. Both inference wires support text, tool calls and results, JSON schemas, and streamed replies through the shared kernel wire.
The World runs no model. Its scenario in handlers/ai-gateway.json scripts text, tools or faults; otherwise answers are labeled deterministic stubs. Catalogue data is a fixed vendor recording. Keys and issued ids are private World bookkeeping. No served call creates vendor-stored resources or emits webhooks.
POST /_twin/app-credentials issues a synthetic bearer key and fills AI_GATEWAY_API_KEY. The door stands in for dashboard key creation. Authentication failures use the gateway error envelope. Model retrieval, embeddings, AI SDK config and credits remain explicit gaps. No numeric per-minute allowance is established by the vendored reference; the live-call budget uses the platform fallback of 30 per minute.
