Models and standalone use
Use a saved hosted agent, or run the same operations with your own standalone provider.
Saved agents use NullProtocol models
Choose an available model in the dashboard. A Space SDK key loads its configuration, authorizes the hosted model and records History. Model availability and credit depend on the team. Changing the saved model applies to the next operation.
Keep tasks narrow and test your real inputs. A small model may handle extraction well and still struggle with policy reasoning. Output schema validation cannot repair a wrong fact.
Use a local standalone agent
Terminal
ollama pull qwen2.5:3b-instructlocal.mjs
import { NullProtocol } from 'nullprotocol';
const agent = await NullProtocol.create({
provider: 'ollama',
model: 'qwen2.5:3b-instruct',
instructions: 'Read only the supplied facts. Do not invent missing values.',
timeout: 120000
});
const result = await agent.extract('Order 42: two blue mugs.', {
orderId: 'number', quantity: 'number'
});
if (!result.success) throw new Error(result.error);
console.log(result.data);Know which limit you are setting
| Setting | What it limits |
|---|---|
| maxTokens | Generated output tokens; default 1000 |
| maxPromptBytes | Serialized request bytes; default 131072, not the model window |
| documentBudget | Saved reference-text characters; default 6000, not all context |
| timeout | Request timeout in milliseconds; default 30000 |
Standalone providers include OpenAI, Anthropic and OpenAI-compatible servers. Their credentials stay in your application. See provider configuration for setup.