开发者找免费AI API的痛苦:每个平台endpoint不同、格式不同、限额不同。LangBot意外解决了这个问题——它不只是个IM机器人框架,还内置了免费API聚合层。
每次找免费API都要翻遍文档,直到发现这个1.6万星的项目

你有没有过这种体验——写了个小项目,想接个AI模型。打开OpenRouter,翻了半小时文档,发现免费额度就那么点。去硅基流动看看,endpoint是https://api.siliconflow.cn/v1,跟OpenAI的不一样。火山引擎Ark又是一套。智谱GLM免费的Flash模型限制一堆。
每个平台的文档长得都不一样。申请key的流程不一样。API格式有微妙差异。限额写在犄角旮旯。
上次我花了一个下午,就为了把三个免费网关都接进项目里。每个都得单独看文档、单独写适配代码、单独处理错误码。搞完之后我跟自己说:下次再也不这么干了。

一、它本来是个IM机器人框架
LangBot是个开源项目,GitHub上16.6k星。初衷是做一个"一个代码库部署到所有IM平台"的机器人框架——Discord、Telegram、Slack、LINE、QQ、微信、企业微信、飞书、钉钉,一共12个平台。
我一开始是冲着它的IM能力去的。
但翻README的时候发现了一行字:内置模型服务,OpenAI兼容格式,支持SiliconFlow、火山引擎Ark、阿里云百炼、ModelScope等网关。
等一下。
这些不都是我之前一个个去翻文档的免费网关吗?
二、原来它是个免费API聚合器
仔细看了它的模型服务层,发现LangBot做了一件很聪明的事:
它把多家LLM网关的API统一封装成了OpenAI兼容格式。
什么意思?就是说不管你用的是SiliconFlow的免费模型、火山引擎Ark的豆包、还是阿里云百炼的通义千问,在LangBot里都是同一个endpoint、同一套参数格式。
你不用再去每个平台查"我的endpoint是什么"、"请求格式有没有差异"、"streaming怎么开"。LangBot全帮你处理了。
具体支持的免费/低价网关:
-
SiliconFlow(硅基流动)——免费额度够测试用,支持Qwen、Llama等开源模型 -
火山引擎Ark——豆包系列模型,有免费调用额度 -
阿里云百炼——通义千问系列,部分模型免费 -
ModelScope——魔搭社区,模型多,部分免费 -
GiteeAI——国内开发者友好 -
302.AI——聚合网关 -
Ollama——本地部署,完全免费
这7个网关加起来,基本覆盖了国内能用的所有免费/低价AI模型渠道。而你只需要在一个地方配置一次。

三、一条命令的事
LangBot的部署比我预期简单太多了。
uvx langbot
一行命令,服务跑起来了。打开http://localhost:5300,Web管理面板直接配置。
在面板里加一个bot、选一个模型提供商、填个API key,三步搞定。它还有个多provider负载均衡——你挂了SiliconFlow的免费额度,自动切换到火山引擎Ark。不用自己写重试逻辑。
对了,它还支持Dify、Coze、n8n这些工作流平台的深度集成。35%的LangBot用户拿Dify当AI后端。也就是说,你甚至可以不直接调API,而是把LangBot当成Dify和IM平台之间的桥梁。
说实话,我本来只想找个能用的免费API。结果LangBot直接给了我一个能跑在Telegram上的完整AI助手。
四、生产级的那些东西
免费归免费,LangBot有几个生产级特性让我觉得它不是玩具:
限流和访问控制——不是把API key裸奔在外面。可以按用户、按群组设限额。
日志追踪——每次对话都有完整记录,token用了多少、延迟多少毫秒,面板上直接看。
插件生态——100多个插件,从定时任务到图片生成到知识库检索,装上就能用。
MCP支持——Model Context Protocol,这意味着它能跟Claude Code、Codex这些AI编程工具对接。
Apache 2.0协议,完全开源,自托管没有功能限制。不想折腾服务器的话,也有LangBot Cloud可以选。
五、说几句实话
LangBot不是完美的。
它的文档偏英文,国内开发者看起来可能有点费劲。Web面板的UI不算好看,功能够用但不精致。有些高级功能(比如多管道架构)的文档写得比较简略,得自己摸索。
但核心问题——"找免费API太麻烦"——它确实解决了。
你不用再花一个下午去翻5个平台的文档。一个LangBot,7个免费网关,统一接口,统一管理。
如果你也在为"免费AI API接入"这件事头疼,去试试uvx langbot。至少省下来的那个下午,你可以干点更有意思的事。
The pain of developers searching for free AI APIs: each platform has different endpoints, formats, and quotas. LangBot unexpectedly solved this problem — it's not just an IM bot framework, but also includes a built-in free API aggregation layer.
Every time I look for a free API, I have to scour through documentation until I found this 16k-starred project.
Have you ever experienced this? You write a small project and want to connect an AI model. You open OpenRouter, browse the docs for half an hour, and find the free quota is limited. You check SiliconFlow, and their endpoint is https://api.siliconflow.cn/v1 — different from OpenAI’s. Then there’s Volcán Engine Ark with another format. Zhipu GLM’s free Flash model has tons of restrictions.
Each platform’s docs are different. The key application process varies. The API formats differ subtly. The quota limits are buried in obscure corners.
Last time, I spent an entire afternoon just integrating three free gateways into my project. I had to read docs, write adapters, and handle error codes separately for each. After that, I vowed: next time, I won’t do it this way again.
Part 1: It’s originally an IM bot framework.
LangBot is an open-source project with 16.6k stars on GitHub. It was designed to deploy bots across all IM platforms — Discord, Telegram, Slack, LINE, QQ, WeChat, Lark, DingTalk — totaling 12 platforms.
I initially got it because of its IM capabilities.
But reading the README, I saw a line: “Built-in model service, OpenAI-compatible format, supporting SiliconFlow, Volcano Engine Ark, Alibaba Cloud Baiqiu, ModelScope, etc.”
Wait a moment.
Aren’t these the free gateways I was manually checking just now?
Part 2: Turns out, it’s a free API aggregator.
Looking closer at its model service layer, I realized LangBot did something clever:
It unified multiple LLM gateways into OpenAI-compatible format.
What does that mean? Whether you use SiliconFlow’s free model, Volcano Engine Ark’s Douyin-bot, or Alibaba Cloud Baiqiu’s Qwen, in LangBot they all use the same endpoint and parameter format.
You no longer need to check “what’s my endpoint?” or “is the request format different?” or “how do I turn on streaming?”. LangBot handles all that for you.
Specifically supported free/low-cost gateways: - SiliconFlow (Silicon-based flow) — free quota for testing, supports Qwen, Llama, etc. - Volcano Engine Ark — Douyin-bot series, with free usage quota. - Alibaba Cloud Baiqiu — Qwen series, partially free. - ModelScope — Modao community, many models, some free. - GiteeAI — developer-friendly in China. - 302.AI — aggregator gateway. - Ollama — local deployment, fully free.
These 7 gateways cover nearly all usable free/low-cost AI model channels in China. You only need to configure once in one place.
Part 3: It’s just one command away.
Deploying LangBot is far simpler than I expected.
One command launches the service. Open http://localhost:5300, and there’s the web admin panel for configuration.
Add a bot, select a model provider, enter an API key — done in three steps. It even has a multi-provider load balancer: if SiliconFlow runs out of quota, it automatically switches to Volcano Engine Ark. No need to write retry logic yourself.
By the way, it integrates deeply with workflows like Dify, Coze, and n8n. 35% of LangBot users use Dify as their AI backend. That means you don’t even have to call APIs directly — you can use LangBot as a bridge between Dify and IM platforms.
Honestly, I just wanted a usable free API. But LangBot gave me a full-featured, Telegram-capable AI assistant.
While it’s free, LangBot has several production-grade features that make it more than a toy: - Rate limiting and access control: API keys are not exposed; limits can be set by user or group. - Log tracing: Full records of each conversation, showing tokens used and latency. - Plugin ecosystem: Over 100 plugins for scheduling, image generation, knowledge retrieval — install and use immediately. - MCP support: Model Context Protocol allows integration with AI coding tools like Claude Code or Codex. - Apache 2.0 licensed, fully open-source, with no functional limits on self-hosting. There’s also LangBot Cloud for those averse to server management.
Part 5: A few honest words.
LangBot isn’t perfect.
The documentation is mostly English, which may be tough for Chinese developers. The web panel UI isn’t beautiful — it’s functional but not polished. Some advanced features (like multi-pipeline architecture) lack detailed docs and require trial-and-error.
But it solves the core problem: “Finding and connecting free AI APIs is too troublesome.”
You no longer have to spend an afternoon browsing 5+ docs. One LangBot connects 7 free gateways with a unified interface.
If you’re struggling with free AI API integration, try `uvx langbot`. At least you’ll save that afternoon for something more interesting.
While it is indeed free, LangBot features several production-grade characteristics that make it more than just a toy:
Rate limiting and access control — APIs are not left wide open. You can set usage limits for individual users or groups.
Logging and tracing — Full conversation histories, including token counts and response times, are available in the dashboard.
Plugin ecosystem — Over 100 plugins exist for tasks ranging from scheduled jobs to image generation to knowledge base retrieval.
MCP support — Integration with the Model Context Protocol allows for interaction with AI programming tools like Claude Code and Codex.
Apache 2.0 license — The software is fully open-source with no restrictions on self-hosting features; LangBot Cloud is also an option for those who prefer not to manage their own server.
五、说几句实话
LangBot不是完美的。
The documentation is primarily in English, which may pose a challenge for Chinese developers. While the web panel is functional, the UI is not particularly stylish. Additionally, the documentation for some advanced features, such as the multi-pipe architecture, is sparse and requires some trial and error to understand.
However, it effectively addresses the core issue of "finding free APIs too cumbersome."
You no longer need to spend an afternoon reading manuals for five different platforms. LangBot provides a single gateway to seven free services with a unified interface and consistent management.
If you are struggling with accessing free AI APIs, consider trying uvx langbot. The time saved in the process could be spent on more productive activities.