GPT-Load Introduces a Self-Hosted Gateway for Multi-Provider AI Access

Summary

GPT-Load is an open-source, self-hosted AI gateway that gives applications one base URL and AccessKey for connecting to multiple providers, model services, compatible relays, and subscription accounts. It supports native OpenAI, Anthropic, Gemini, and Mistral interfaces, along with OpenAI-compatible endpoints for chat, responses, images, embeddings, and reranking. Operators can manage API keys, OAuth-based subscription credentials, providers, models, groups, and routing policies through a web console. The gateway provides weighted scheduling, retries, cooldowns, blacklisting, session affinity, health handling, and failure isolation across credentials. It also exposes request logs, traffic and health views, usage statistics, cache-hit information, and cost estimates, with data stored in SQLite by default or in MySQL and PostgreSQL through a database connection string. Credentials are encrypted locally, so the encryption key must remain with the database during backup or migration. GPT-Load 2.0 is a complete rewrite and cannot directly open or migrate 1.x data, requiring a separate deployment and verification before traffic is moved. The project can run with Docker Compose or native binaries, supports Windows service installation, and is released under the MIT License.