Spillway is an open-source, self-hosted AI gateway for teams that use cloud and local language models. It runs as a single Docker container with an admin UI and SQLite storage.
Clients connect with the OpenAI, Anthropic or Ollama request formats, including the OpenAI Responses API that Codex uses, and Spillway translates between them, with streaming and tool calls. Each person, device or agent gets its own key with daily and monthly spend limits. Team budgets and routing rules can send requests to a local model through Ollama when a budget threshold is reached, when a cloud provider fails, or outside working hours.
For coding agents, Spillway checks whether the local model can stand in for a cloud one (context length, tool calls, speed) and can give an Ollama model the longer context they need in one click. Claude Code users see what their key has spent and where the next request goes in Claude Code's status line.
The request log masks personal data such as emails, card numbers and IBANs, and prompts with sensitive data can be kept away from cloud models. Single sign-on through OpenID Connect (Google Workspace, Microsoft Entra ID, Keycloak) is included. Licensed AGPL-3.0.
No comments or reviews, maybe you want to be first?