GPT-Load

Open source

Self-hosted AI gateway written in Go that puts many model channels and credentials behind one base URL, under the MIT licence.

gpt-load.com
GPT-Load homepage screenshot
GitHub stars
7.2k
Last 30 days
+135(+2%)
Last commit
yesterday
Repository age
17 months
Version
v2.0.0-rc.63
Licence
MIT
Self-hosted
Yes

About GPT-Load

GPT-Load is a self-hosted gateway for AI model APIs. Instead of each application managing its own keys and providers, the gateway sits in front and exposes one base URL. The site describes dozens of credentials spread over more than thirty channels, handled by a single Go binary that stores its data in SQLite, MySQL or PostgreSQL.

The gateway takes over scheduling, retries, cooldown and usage accounting. API keys and subscription accounts for Codex, Claude, Antigravity and Grok share one pool with the same health-isolation behaviour. Weighted rotation and session affinity are also named as ways failures are routed away.

Clients keep their own protocol: OpenAI Chat Completions, OpenAI Responses, Anthropic Messages and Gemini are supported, with each channel declaring a native or converted route. Applications are told to configure only a base URL and an AccessKey. Credentials are encrypted locally and requests go to the upstreams you choose. Version 2.0 is current on the site, which is available in English, Chinese and Japanese.

Key features

  • Many channels and credentials behind one base URL
  • API keys and subscription accounts in one scheduling pool
  • Supports four client protocols
  • Clients only need a new base URL and AccessKey
  • Stores data in SQLite, MySQL or PostgreSQL

Good fit for

  • Pooling several API keys for one provider
  • Giving applications a single endpoint for many models
  • Tracking usage across providers
Built with
Go tools
Tags
ai gateway
llm proxy
api keys
openai compatible
self-hosted
go

GPT-Load: questions and answers

What is GPT-Load used for?
GPT-Load is a self-hosted AI gateway written in Go that puts many model channels and credentials behind one base URL, under the MIT licence. It is a good fit for pooling several API keys for one provider, giving applications a single endpoint for many models and tracking usage across providers.
Is GPT-Load open source?
Yes. GPT-Load is open source under the MIT licence. Its source code is on GitHub at tbphp/gpt-load and is written mainly in Go.
Is GPT-Load free?
Yes. GPT-Load is open source, so the software itself is free to use.
Can I self-host GPT-Load?
Yes. GPT-Load can be self-hosted on your own server or infrastructure.
What are some alternatives to GPT-Load?
Similar open-source tools in the AI Infrastructure category include Obot, beta9 and CoAI. SaaS products in the same category include Polyrouter, AI Stats and Audn Corporation.
Is GPT-Load actively maintained?
Yes. The most recent commit to GPT-Load was on 12 November 2026, and the latest release is v2.0.0-rc.63, published on 8 November 2026. The project has 7.2k stars on GitHub, with 135 added in the last 30 days.

Open-source alternatives to GPT-Load

See all

SaaS alternatives to GPT-Load

See all