AI in your app

​

What you can build

​

How it fits together

1. Connect a provider (OpenAI, Anthropic, Google, Mistral, Groq, or your own server) in AI → Providers. Your key is stored on the server and never shown again.
2. Add models with names your app uses, such as chat and embed. You can change which real model sits behind a name later without touching your code.
3. Use them from your app, from flows, or through agents. Who can use each model, agent and knowledge base is an access rule you choose, the same way tables and files work.

​

What stays in your control

• Your provider keys never reach a browser or an app. Your app only ever holds a Pawabase key.
• Agents act as the person who asked. By default an agent can only read and change what that person could, so it can never show someone a row they could not open themselves.
• Changes wait for a person unless you say otherwise. An agent that wants to delete, update or send something stops and asks.
• Everything is on the record. Each model call and agent run keeps its tokens, time and cost, and each agent step is listed, so you can see what happened and why.
• Limits you set. Cap how many steps, how long and how much an agent may spend on one run.

​

A first model in two minutes

const pawabase = createClient({ url, apiKey: "pb_sk_…", environment: "production" });
const answer = await pawabase.ai.chat("chat", "Write a one-line description for a ceramic mug.");

pb = Pawabase(url, "pb_sk_…", environment="production")
answer = pb.ai.chat("chat", "Write a one-line description for a ceramic mug.")