GoModel

GoModel is a self-hosted and open-source AI Gateway written in Go (that's why it's called GoModel).

It puts one fast, stable layer between your applications and 22 AI providers. Keep using the OpenAI or Anthropic SDK, change the base URL, and route requests to OpenAI, Anthropic, Gemini, Azure OpenAI, Amazon Bedrock, Groq, Ollama, vLLM, and many more.

GoModel is an alternative to LiteLLM, whose PyPI packages were recently compromised in a supply-chain attack, and Portkey, whose public gateway repository is no longer actively maintained.

What you get:

• One API for 22 providers - switch providers and models without rewriting application code.

• Virtual models and aliases - expose stable names such as smart-chat while changing the real model behind them whenever you want.

• Realtime logs - you can see what's going on immediately.

• Load balancing and failover - spread traffic across models, retry temporary failures, avoid unhealthy providers with circuit breakers, and automatically move requests to configured backups.

• Exact and semantic caching - return repeated or similar responses without paying the provider again. Semantic caching supports Qdrant, pgvector, Pinecone, and Weaviate.

• Usage and cost tracking - see tokens and dollars for every request, team, tenant, model, and label. Override model prices when your contract differs from public pricing.

• Budgets and rate limits - enforce spending limits before requests are sent, then control request volume, token usage, and concurrency at the user path, provider, or model level.

• Scoped workflows and guardrails - apply different caching, audit, budget, failover, and request-processing rules to different teams, applications, providers, or models.

• Virtual API keys and user paths - give applications managed keys instead of provider credentials, restrict model access, attach labels, and attribute usage to the right customer or team.

• Full observability - inspect live requests, provider attempts, cache results, costs, and audit logs in the embedded dashboard. Export Prometheus metrics when you need them.

• More than chat completions - use the Responses API, Anthropic Messages API, embeddings, files, batches, audio, realtime speech, and provider-native passthrough routes through the same gateway.

• An MCP gateway - combine multiple MCP servers behind one authenticated endpoint with namespaced tools, usage tracking, and audit logs.

• Simple operations - run one small Go binary with an embedded admin dashboard. Start with SQLite, then move to PostgreSQL or MongoDB without changing the gateway.

GoModel saves you money and reduce your AI bills. Duplicate prompts can be served from cache, cheaper models can handle suitable traffic, every expense is attributable and budgets stop surprises before they happen.