# aisuite review

> aisuite is a lightweight Python library offering a unified Chat Completions API across LLM providers plus an Agents API with tools and toolkits.

Devtoolwatch Board, the week of Oct 25–31, 2026. Canonical page: https://index.devtoolwatch.com/tools/aisuite/

## TL;DR

aisuite is a lightweight Python library for building with LLMs. aisuite gives Python developers one OpenAI-style API across providers such as OpenAI, Anthropic, Google and Ollama, where you swap providers by changing one string. On top of that, aisuite adds an Agents API with Python-function tools, sandboxed toolkits, MCP support and tool policies.

## aisuite at a glance

- **Category:** [Model SDKs and gateways](https://index.devtoolwatch.com/categories/model-sdks-and-gateways/)
- **Maintained by:** Andrew Ng (individual, [@andrewyng](https://github.com/andrewyng) on GitHub)
- **License:** MIT
- **Language:** Python, TypeScript, HTML
- **Repository:** [github.com/andrewyng/aisuite](https://github.com/andrewyng/aisuite)
- **PyPI package:** [aisuite](https://pypi.org/project/aisuite/) 0.2.0
- **Latest release:** [aisuite 0.2.0](https://pypi.org/project/aisuite/) on PyPI, Sep 18, 2026
- **First published:** Jun 30, 2024

## What aisuite does

- aisuite provides a unified, OpenAI-style chat API for OpenAI, Anthropic, Google, Mistral, Hugging Face, AWS, Cohere, Ollama, OpenRouter, Requesty and more.
- aisuite streams OpenAI-shaped chunks with stream=True from supporting providers, with an async acreate variant.
- aisuite turns plain Python functions into tools, generating schemas, executing calls and feeding results back when max_turns is set.
- aisuite's Agents API runs declared agents with a Runner and attaches sandboxed toolkits for files, git and shell.
- aisuite supports tool policies such as RequireApprovalPolicy and allow/deny lists to govern which tool calls run.
- aisuite persists and resumes agent runs with in-memory, file or Postgres state stores, and captures artifacts and traces.
- aisuite natively supports the Model Context Protocol, so any MCP server's tools can be passed to a model.

### How it works

You install aisuite from PyPI, create an ai.Client(), and call client.chat.completions.create with a model string in provider:model format. aisuite routes the call to the right provider SDK with the right parameters. Provider support comes from adapters discovered by a naming convention, such as an OpenaiProvider class in openai_provider.py. Agents run through Agent and Runner, and MCP servers can be attached as tools.

## How to install aisuite

**Base package (no provider SDKs)**

```
pip install aisuite
```

**With all provider SDKs**

```
pip install 'aisuite[all]'
```

**With MCP support**

```
pip install 'aisuite[mcp]'
```

### Requirements

- Python 3.10 or later, below 4.0.
- API keys for each LLM provider you call.
- Provider SDKs installed via extras; the base package includes none.

### First steps

1. Install aisuite, adding the SDK extras for the providers you plan to call, such as 'aisuite[anthropic]'.
2. Set up API keys for each provider, following the Chat Completions quickstart in docs/chat-completions-quickstart.md.
3. Create a client with ai.Client() and call client.chat.completions.create using a model string like "openai:gpt-4o".
4. Pass Python functions in tools with max_turns to let aisuite run the tool-calling loop for you.

## Who aisuite is for

- Python developers who want to call several LLM providers through one OpenAI-style interface.
- Developers comparing models by swapping the provider string without rewriting code.
- Teams building tool-using agents with approval policies, persistent state and MCP servers.

### Limitations

- aisuite is a Python library; the README shows no SDK for other languages.
- aisuite streaming is manual tool calling and cannot be combined with max_turns.
- aisuite streaming works only with supporting providers: OpenAI, Anthropic, Ollama and OpenAI-compatible endpoints.
- The OpenWorker desktop app built on aisuite is developed in a separate repository, not in aisuite itself.

## Usage data (the week of Oct 25–31, 2026)

- Board rank: #243 overall, #13 in Model SDKs and gateways, momentum score 39
- Downloads that week: 21k (+105% vs 4-week average)
- New GitHub stars per week (last 12 weeks): Aug 9: 40; Aug 16: 36; Aug 23: 86; Aug 30: 39; Sep 6: 30; Sep 13: 29; Sep 20: 33; Sep 27: 21; Oct 4: 12; Oct 11: 26; Oct 18: 21; Oct 25: 18

## aisuite alternatives

| Rank | Tool | Stars | Downloads/wk |
|---|---|---|---|
| #1 | [thinking-orbs](https://index.devtoolwatch.com/tools/jakubantalik-thinking-orbs/) | 3.9k | 1.4M |
| #2 | [Headroom](https://index.devtoolwatch.com/tools/headroomlabs-ai-headroom/) | 78.5k | 125.8k |
| #3 | [LiteLLM](https://index.devtoolwatch.com/tools/litellm/) | 62.5k | 23.1M |
| #4 | [Honcho](https://index.devtoolwatch.com/tools/plastic-labs-honcho/) | 7.9k | – |
| #5 | [Anthropic SDK](https://index.devtoolwatch.com/tools/anthropic-sdk/) | 4k | 111M |
| #6 | [Hugging Face Hub](https://index.devtoolwatch.com/tools/hf-hub/) | 4k | 68.3M |

## aisuite FAQ

### Is aisuite free and open source?

Yes. aisuite's source code is on GitHub (andrewyng/aisuite) under the MIT license, so you can use it for free.

### How do I install aisuite?

Run `pip install aisuite`. The aisuite package needs Python 3.10 or newer, below 4.0. The install section above has the other options from the project's README.

### Is aisuite still maintained?

Yes. The andrewyng/aisuite repository was last updated on Sep 18, 2026, and the latest release, aisuite 0.2.0 on PyPI, came out on Sep 18, 2026.

### How popular is aisuite?

aisuite has 16.4k GitHub stars as of Nov 1, 2026. It gained 18 new stars in the week of Oct 25–31, 2026, against a 12-week average of 33 a week. Its main package had 21k downloads that week, +105% against its own 4-week average. That made it #13 among LLM SDKs and gateways on the Devtoolwatch Board.

### What are the best alternatives to aisuite?

Other LLM SDKs and gateways on the Devtoolwatch Board in the week of Oct 25–31, 2026, by momentum: thinking-orbs (#1), Headroom (#2), LiteLLM (#3), Honcho (#4), Anthropic SDK (#5). The alternatives table above has their numbers.

### What is the aisuite library?

aisuite is a lightweight Python library for building with LLMs. aisuite has two layers: a unified Chat Completions API across providers, and an Agents API with tools, toolkits and MCP support on top.

### How do I switch between LLM providers in aisuite?

aisuite selects the provider from the model string in provider:model format, such as "openai:gpt-4o" or "anthropic:claude-3-5-sonnet-20240620". Changing that one string routes the same call to a different provider.

### Does aisuite support MCP servers?

Yes. aisuite natively supports the Model Context Protocol after installing the mcp extra. You can pass an MCP server as a tool entry, or use the explicit MCPClient for reusable connections, security filters and tool prefixing.

### What is the relationship between aisuite and OpenWorker?

OpenWorker is a desktop AI coworker built on aisuite. It runs on macOS and Windows. OpenWorker development has moved to its own repository, andrewyng/openworker, and a historical snapshot remains in openworker-archive/.

### Can I add a new provider to aisuite?

Yes. aisuite adds providers through a lightweight adapter. You create a module file named like openai_provider.py with a capitalized class such as OpenaiProvider extending BaseProvider. The naming convention enables automatic loading.

## Sources

1. [andrewyng/aisuite on GitHub](https://github.com/andrewyng/aisuite)
2. [aisuite README](https://github.com/andrewyng/aisuite/blob/main/README.md)
3. [aisuite on PyPI](https://pypi.org/project/aisuite/)
4. [Release v0.1.3 (Jul 20, 2026)](https://github.com/andrewyng/aisuite/releases/tag/v0.1.3)
5. [Release app-v0.1.1 (Jun 11, 2026)](https://github.com/andrewyng/aisuite/releases/tag/app-v0.1.1)
6. [Release app-v0.1.0 (Jun 11, 2026)](https://github.com/andrewyng/aisuite/releases/tag/app-v0.1.0)
7. [Hacker News: Andrew Ng ships Open CoWorker desktop AI agent](https://news.ycombinator.com/item?id=48562975)
8. [Devtoolwatch Board, the week of Oct 25–31, 2026](https://index.devtoolwatch.com/)