ai-gateway-vivid-fish

What problem does it solve?

It solves the problem of needing multiple AI providers and local backends by giving you a single OpenAI/Anthropic/Ollama-compatible endpoint that can route chat, vision, embeddings, TTS/STT, image generation/editing, audio/video jobs, and model/voice discovery.

Core Features & Use Cases

• Unified gateway routing: Use one base URL to access local Tabby/Kobold LLMs, Voxtral TTS, Parakeet STT, Nomic embeddings, and ComfyUI for image generation and edits.
• OpenAI- and Anthropic-shaped APIs: Call /openai/v1/*, /anthropic/v1/messages, and /ollama/api/* with consistent request/response behavior and discovery endpoints.
• Durable media jobs for long tasks: Submit async image/audio/video jobs and poll for completion to fetch signed output handles without keeping long HTTP requests open.
• Vision-aware chat completions: Automatically routes image-bearing chat requests to a vision-capable backend or rejects with a clear error when vision support is unavailable.
• Secure app/admin key separation: Use AI_GATEWAY_API_KEY (opk_...) for runtime calls and reserve oak_... admin keys for admin/control-plane routes.

Quick Start

Ask the AI to call the Vivid Fish gateway using your app bearer token to list available models, then run a chat completion on model: "default".