
Give your AI
the right context.
Give each agent its rules, context and knowledge. Call it from your app, change its instructions without a deploy, and see what went into every result.
JavaScript, TypeScript & Python. HTTP API included.- days
- 30
A small model for a focused task.
A bigger one when you need it.

Turn everyday text
into useful data.
Extract structured data, summarize messages, check answers and choose the next action. Four operations for the work inside your app.
Extract
Turn unstructured text into the fields your app needs.
Order 1042 arrived with a cracked mug. Can I get a replacement?
- orderId
- 1042
- category
- damaged_item
- nextStep
- request_photo

Update prompts.
Keep the code.
Manage agent instructions, reference texts and memory in the dashboard. Your app picks up the changes on the next operation, without a redeploy.
- Share rules across agents
- Keep customer data separate
- Save facts your agents can use
Unused purchases can be returned within
Keep the original packaging. Gift cards are not refundable.
“Within how many days can I return it?”
{ days: 30 }Jamie prefers a replacement over a refund.
Find the answer
inside your documents.
Add guides, policies and product docs. Connect the sources an agent needs, then search them from any operation.
One shared library. The right sources for each agent.
How long is the return window?
Unused purchases can be returned within 30 days of delivery. Keep the original packaging. Gift cards are not refundable.
{ "days": 30 }Manage sources in the dashboard or from code. Keep the exact passages with every result.
Explore Knowledge
Debug an answer.
See the context.
Inspect the input, output, instructions and Knowledge passages behind each operation. See which rules were in effect when an answer went wrong.
| Operation | Result | Tokens in / out |
|---|---|---|
| Completed | 422 / 42 | |
| Fail | 318 / 34 | |
| Completed | 286 / 58 |
ExtractCompleted
- nextStep
- request_photo
Order 1042 arrived with a cracked mug. Can I get a replacement?
Context, captured
Read the exact version an operation ran with.
Token usage, tracked
See model usage for each operation and each agent.
Actions, visible
Follow a decision through to its recorded execution.

Fits into the app
you already have.
Load an agent by its ID. Choose an operation and get structured results in JavaScript, TypeScript or Python.
Keep your app’s workflow. Change the agent’s model, rules and sources in one place.
Read the documentationimport { NullProtocol } from 'nullprotocol';
const agent = await NullProtocol.load({
key: process.env.NULLPROTOCOL_KEY,
agentId: 'support'
});
const result = await agent.extract(
'How long is the return window?',
{ days: 'number' },
{ knowledge: true }
);
if (result.success) console.log(result.data);{ "days": 30 }npm install nullprotocol
Put your first agent to work.
Start with one task. Give it the context to do it well.