llm-proxy-tk puts enterprise model access behind an intranet gateway

llm-proxy-tk is an open-source FastAPI gateway designed to bring enterprise LLM usage under centralized intranet control. Instead of employees sharing vendor API keys or using public chat tools directly, teams route model calls through a self-hosted access layer that issues independent keys, authenticates users, meters usage, and connects to upstream providers such as DeepSeek, Claude, Ollama, vLLM, Zhipu, and OpenAI-compatible endpoints.

The gateway targets common governance problems for small and mid-sized teams, including leaked credentials, sensitive data leaving the corporate perimeter, and hard-to-manage integrations across multiple model vendors. Administrators can set key expiration, reset compromised access, audit call volume and token use, and view usage by department or application. Local Ollama or vLLM deployments can be added as upstream providers so finance, legal, and core code teams can keep prompts and outputs inside the corporate network.

The project stays deliberately lightweight, with Docker or bare-metal deployment, SQLite storage, offline installation support, and a built-in chat interface. It does not include billing, content moderation, or high-availability clustering. Ollama’s v0.33.0 release points to a similar shift at the access layer, adding a local proxy for Claude Desktop that can route labels such as Opus 5 to open models including Kimi K3.