GoModel - A high-performance AI gateway written in Go

GoModel is a lightweight AI gateway distributed as a single Go binary. It bundles 11 providers including OpenAI, Anthropic, Gemini, Groq, and xAI into a single OpenAI-compatible API.

Just drop API keys into a Docker container as environment variables and it automatically detects providers, ready to use immediately. Supported endpoints cover a wide range: /v1/chat/completions (streaming), /v1/responses, /v1/embeddings, /v1/models, /v1/files, and /v1/batches.

Features the gateway can't handle are passed straight through to the upstream via native /p/{provider}/... passthrough. The response cache is structured in 2 layers.

• Layer 1: an exact-match cache based on request body hashing, offering sub-millisecond lookups
• Layer 2: a semantic cache based on embedding vector KNN, claiming a 60-70% hit rate for highly repetitive workloads
• Qdrant, pgvector, Pinecone, and Weaviate are supported as vector backends.

Operational features are also included. The dashboard shows token usage, cost, audit logs, provider status, and model overrides, and Prometheus metrics, audit logging, and a guardrails pipeline can be enabled via environment variables. Authentication is handled via GOMODEL_MASTER_KEY, and storage can be chosen from SQLite, PostgreSQL, or MongoDB.

Multiple instances of the same provider can also be registered using suffixed environment variables. The 0.2.0 roadmap includes intelligent routing, addition of DeepSeek V3 and Cohere, per-user_path budget management, and cluster mode. It's licensed under MIT.