The anti-dashboard: route LLMs through a YAML file
uni-api exists because one/new-api is overkill for personal use—just write a YAML file to unify dozens of LLM backends into a single OpenAI-compatible endpoint.

What it does
uni-api is an API gateway that normalizes calls from OpenAI, Anthropic, Gemini, Vertex, Azure, AWS, Groq, and others into the OpenAI request/response format. You define providers and API keys in a single api.yaml, and the proxy handles routing, retries, and load balancing without shipping a frontend. It exposes the standard OpenAI endpoints—chat completions, embeddings, image generation, audio, and moderation—so downstream tools don’t need to know which backend is actually answering.
The interesting bit
The project treats “no frontend” as a feature, not a missing feature. Instead of click-ops, you configure weighted load balancing, channel cooling after failures, per-model timeouts, and rate limits directly in YAML. It even exploits Vertex AI’s regional endpoints to multiply your Gemini and Claude concurrency without extra configuration.
Key highlights
- Supports native tool use and image recognition across OpenAI, Anthropic, Gemini, Vertex, Azure, AWS, and xAI, then presents them through standard OpenAI endpoints.
- Four load-balancing strategies: channel-level weighted, sequential round-robin, API key polling, and automatic Vertex regional distribution.
- Model renaming and request-body parameter overrides let you expose simplified model names or inject provider-specific flags without touching client code.
- Built-in moderation review can reject inappropriate messages before they reach the backend, reducing the odds of your provider API key getting banned.
- Rate limiting supports per-minute, hour, day, month, and year quotas.
Caveats
- Built explicitly for personal use; commercial features like multi-tenant billing are out of scope by design.
- No frontend means every change requires editing
api.yaml—there is no admin panel for on-the-fly tweaks.
Verdict
If you run personal LLM experiments and want one OpenAI-shaped endpoint that load-balances across your scattered API keys, uni-api is a leaner alternative to full-featured dashboards. If you need user management, billing, or a GUI, keep looking—this project deliberately isn’t that.
Frequently asked
- What is yym68686/uni-api?
- uni-api exists because one/new-api is overkill for personal use—just write a YAML file to unify dozens of LLM backends into a single OpenAI-compatible endpoint.
- Is uni-api open source?
- Yes — yym68686/uni-api is open source, released under the Apache-2.0 license.
- What language is uni-api written in?
- yym68686/uni-api is primarily written in Python.
- How popular is uni-api?
- yym68686/uni-api has 1.2k stars on GitHub.
- Where can I find uni-api?
- yym68686/uni-api is on GitHub at https://github.com/yym68686/uni-api.