AI gateways (retired)

Fountain used to sit behind an AI gateway — LiteLLM, Portkey, Kong AI Gateway, Cloudflare AI Gateway — as just another OpenAI-compatible upstream: point the gateway's api_base at https://your-fountain/v1, name a Fountain agent as the model, and every client already behind that gateway could reach it. The endpoints that made that work are gone, removed in the release that carries ADR 0057. The examples/litellm-gateway example went with them.

This page stays because the URL is in other people's notes.

What a gateway sees now

A 404 from https://your-fountain/v1/chat/completions, with the ordinary body {"errors": {"detail": "Not Found"}} — not a gateway-shaped error, and not a model that has stopped responding. A gateway configured this way will report the upstream as failing; remove the Fountain entry from its config.

What has no replacement

Putting Fountain behind a gateway with no code. Gateways speak the OpenAI dialect by definition, Fountain no longer does, and the native conversation API is not a chat-completions endpoint — a conversation is created, prompted and followed over several calls rather than answered in one. No gateway can be configured into that shape.

If the gateway was giving you key management, spend tracking or rate limiting across teams, Fountain has its own: API keys, credits and billing and per-key rate limits.

What you can still do

Call the conversation API from your own code, or through the TypeScript, Python, Elixir or Swift SDK. Read Plug into Fountain for the routes in.

Gateways in the other direction are unaffected: Fountain still runs agents against OpenAI, Anthropic and the rest, and a deployment can still put its own egress through a proxy. That is the egress credential broker (secrets), and it shares nothing with the retired inbound dialect.