API Claws by Buda

Build self-improving agent harness with one API

API

Developer Tools

Development

Visit WebsiteSee on Product HuntTwitter ⧉

@mrkelly @kelvinpoon @seey @doriiiiis_ How easy is it to connect a cloud agent to an existing app using your API? This looks useful for developers who want to save time.

Hey Product Hunt! 👋 I'm one of the makers behind API Claws by Buda.

When building an AI-powered app or smart hardware product, a simple AI completion API often isn't enough.

What we really need is an AI agent that can understand specific workflows, remember context, and handle complex, multi-step tasks.

But building an agent from scratch means putting together a lot of infrastructure: memory, model routing, file systems, runtime management, and more. That's a lot of work before you can even focus on what makes your product unique.

So we asked ourselves: What if adding an AI agent to your product could be as simple as calling an API?

That's why we built API Claws.

With API Claws, developers can create their own cloud-native agents, define their roles, build their knowledge base, and customize their tools and skills. Then connect those agents to any app or smart hardware through an API, without worrying about the underlying agent infrastructure.

Our goal is simple: let developers focus on building great products, while we take care of the agent infrastructure behind them.

Hey Product Hunt, I'm Seeyou Chan. I’m one of the makers of API Claws.

As a developer, I think it’s especially useful for Micro-Apps and Smart Hardware: your product can use a cloud agent through an API, without building and maintaining the full agent infrastructure yourself.

Each cloud agent has an isolated Linux runtime and its own Drive. It can run code, work with files, and reuse information saved in its Drive across sessions.

The smartwatch example in our video shows this division of work: the device collects workout data and displays results, while the cloud agent uses saved training history and user feedback to adjust the next training plan.

That’s what interests me: you can focus on your product’s interactions and workflows, while we host the agent runtime and persistent storage.

If you’re building an app or a hardware product, what task would you want a cloud agent to handle for your users?

As a founder building an AI startup, the infrastructure tax is real. You spend weeks on sessions, memory, routing, observability before you ship anything users actually see. API Claws looks like it's trying to collapse that into an API call, which is the right instinct. Drive memory plus managed runtime plus model routing in one place is the part I'd want to test first. Curious how it handles retries and long-running sessions when things go sideways, that's usually where the seams show.

Hey Product Hunt 👋

We built API Claws because product teams keep hitting the same wall: a model API can answer a question, but a product needs an agent that remembers, keeps working on its own, can be watched, and can safely sit in front of real users.

API Claws is Buda's hosted Agent API. Instead of another completion endpoint, you get the agent layer already built:

- API Agents: each with its own role, instructions, and its own Drive
- Drive memory: put manuals, policies, and customer context into files once, and every run reuses them instead of a giant prompt. Choose fast block storage or standard object storage per agent
- Models already wired in: Claude, GPT, Gemini, DeepSeek, plus GPT Image, Nano Banana, and Whisper, or let Auto pick
- Sessions and runs: async by default. Stream or poll, send attachments, keep full history, cancel any run
- Ship it anywhere: short-lived embed URLs so your frontend never holds your API key, scheduled tasks for unattended work, messaging channels, and a full OpenAPI spec

The happy path is small: enable API Claws in the Developer Center, create an agent, write files to its Drive, start a session, then stream or poll the reply. We open-sourced a quickstart that does exactly that:
npx skills add buda-ai/api-claws

Where it fits: smart hardware (the device handles input and output, the agent lives in the cloud), SaaS copilots, support widgets, browser extensions, vertical apps, and internal tools.

Usage is credit-based. No GPUs, no inference servers, no runtime to operate.

Question for builders: if you added an agent to your product tomorrow, which layer would you least want to build yourself: memory, model routing, sessions, embeds, or scheduling?