mrkeyoor.com_
Mon 03 Aug 17:05 UTC
AI Toolsevaluationupdated 03 Aug 2026

new-api

New API is a self-hosted gateway that unifies dozens of different Large Language Model (LLM) APIs into a single, OpenAI-compatible endpoint. It solves the problem of managing multiple API keys, tracking usage, and billing users or departments for their AI consumption.

Verdict

New API is a powerful, feature-packed LLM gateway that excels at management and monetization. If you're running a business that resells API access or an organization struggling to wrangle AI costs, it's one of the best open-source tools available. However, its rapid development, high volume of open issues, and restrictive AGPL license mean you should be prepared for a hands-on, potentially bumpy ride.

Setup4/5Basic Docker setup is very fast, but production config takes work.
Docs3/5Has dedicated documentation, but complexity requires it to be excellent.
Community4/5Extremely popular and active, but has a high number of open issues.
Maturity2/5Still in a release candidate phase; not yet a stable 1.0 release.

Who it’s for

  • Startups or developers looking to resell or monetize access to various LLM APIs.
  • Companies that need a central point of control to manage API keys, monitor costs, and enforce usage policies across different teams.
  • Power users and developers building applications that need to switch between different AI models (like from OpenAI to Claude to a local model) without rewriting client-side code.

Who it’s NOT for

  • Individuals who only need a simple graphical interface to chat with an LLM; this is backend infrastructure, not a chat client.
  • Teams looking for a simple, lightweight proxy without features like billing and user management; the complexity here is overkill for that use case.
  • Businesses that are unable to comply with the AGPL-3.0 license, which requires modifications to be open-sourced if the software is offered as a service over a network.

Setup reality

The README's recommendation of Docker Compose is spot-on and makes the initial setup surprisingly simple. For a basic instance using the default SQLite database, you can be up and running in minutes with just a git clone and docker-compose up. However, this simplicity hides the deeper configuration required for a production environment, such as setting up MySQL, configuring payment gateways like Stripe, and integrating authentication providers like OIDC. Expect the initial launch to be easy, but budget significant time for tailoring it to your specific production needs.

In the rapidly expanding universe of AI models, developers face a chaotic reality: every provider has its own API, its own authentication, and its own billing system. Juggling keys for OpenAI, Azure, Claude, and a half-dozen others is a recipe for messy code and runaway costs. New API is an ambitious open-source project that aims to bring order to this chaos. It acts as a central gateway, a single point of entry that lets you manage, monitor, and monetize access to a wide array of LLM services.

At its core, New API is a sophisticated proxy server. You feed it your collection of API keys from various providers, and it exposes a single, unified API endpoint that is compatible with OpenAI's standards. This is its killer feature. Your applications only need to speak one language—the OpenAI dialect—and New API handles the translation and routing to the correct upstream service. This dramatically simplifies development and makes your applications model-agnostic.

The Business-in-a-Box Gateway

Where New API truly distinguishes itself from simpler proxies is in its extensive suite of management and billing features. It’s built with the assumption that you're not just using AI, but managing its use for others, whether they're internal teams or paying customers. The feature list reads like a checklist for a SaaS startup:

This comprehensive feature set makes New API a powerful tool for organizations. A company's platform engineering team could deploy it to give developers self-serve access to AI models while maintaining central control over budgets and security. A startup could use it as the entire backend for a business that resells curated access to premium AI models.

The Bleeding Edge Has Sharp Corners

While powerful, New API is not a tool you can set and forget. Its biggest weakness is its maturity. The latest version, v1.0.0-rc.23, is a release candidate, not a final, stable release. This signals that the project is still in a state of flux. With over 1,100 open issues on GitHub, it's clear that the rapid pace of development is creating a long tail of bugs, feature requests, and necessary refinements. This is the classic trade-off: you get the latest features, but you also take on the risk of instability.

Another critical consideration is the AGPL-3.0 license. This is a strong "copyleft" license. If you use New API as-is, you're fine. But if you modify its source code and then offer it as a service to others over a network, you are legally obligated to make your modified source code publicly available. For many businesses looking to build proprietary technology on top of an open-source base, this is a non-starter.

The project's documentation is external, which is a good practice, but the sheer number of features means the quality of that documentation is paramount. An administrator will need to spend considerable time learning the system's intricacies to configure billing policies, manage user permissions, and troubleshoot routing issues correctly.

Community and Ecosystem

With over 44,000 stars on GitHub and a release just two days ago (as of this review), New API's community is undeniably massive and active. This is a double-edged sword. The popularity ensures a steady stream of feedback and a high likelihood that someone else has encountered any problem you might face. However, the high issue count suggests the maintainers may be struggling to keep up. This is a project for those comfortable navigating active, and sometimes chaotic, open-source communities.

It's also important to note its heritage as a fork or evolution of One API. New API maintains database compatibility with this predecessor, offering an upgrade path for existing One API users who need more advanced features. This positions One API as the more stable, conservative choice, and New API as the feature-rich, progressive option.

In a real-world stack, New API sits between your applications and the AI model providers. It becomes a critical piece of infrastructure, centralizing logic that would otherwise be scattered across multiple services. For any serious deployment, it should be run in a high-availability configuration and backed by a robust database like MySQL, not the default SQLite.

Alternatives

ProjectWhat it isPick it when
LiteLLMA Python-based package to call 100+ LLMs using a consistent, OpenAI-compatible format.you need the absolute widest range of model support or want to integrate LLM routing directly into your Python application as a library, not just a standalone service.
One APIThe original project that New API is based on, providing a stable, unified API interface for LLMs.you want a more stable, less feature-heavy version of the same core idea. It's a better choice if you don't need New API's advanced billing and authentication features.
LibreChatA full-featured, self-hosted chat UI that can connect to multiple AI backends.your primary need is a user-facing chat application, not a backend API gateway for programmatic use. It includes some proxying capabilities but is client-focused.

Sources

  1. QuantumNous/new-api GitHub Repository
  2. New API Homepage