Chat tiers and managed models
Pick how fast or how deep. DocSlurp picks the model.
You choose a category: a chat tier, a processing quality, an embedding or vision profile. DocSlurp manages the models behind each one and keeps them current as models and prices change, so your integration never pins a model that is retired or repriced.
On this page
Chat tiers
tier | Use it for |
|---|---|
fastest (default) | Quick questions and everyday answers, at the lowest cost. |
balanced | More depth for analysis and comparisons. |
smart | Complex reasoning and detailed answers. |
Add reasoning (none, the default and fastest, then minimal, low, medium, high) for long multi-part questions. Responses report the tier that answered in debug.model. GET /v1/billing/prices lists each tier's credit price.
Pick a tier
Send tier on /v1/chat, /v1/chat/stream, the MCP chat tool and SDK chat(), or --tier with the CLI. With an OpenAI client, keep model as the project and send the tier in the docslurp extension (extra_body in the Python SDK).
import { createDocSlurpClient } from '@docslurp/sdk';
const client = createDocSlurpClient();
const answer = await client.chat({
q: 'What changed in the 2026 travel policy?',
tier: 'smart',
});
console.log(answer.answer, answer.debug?.model); // "smart"curl https://docslurp.io/v1/chat \
-H "Authorization: Bearer $DOCSLURP_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"q": "What changed in the 2026 travel policy?",
"tier": "smart",
"reasoning": "low"
}'import OpenAI from 'openai';
const client = new OpenAI({ baseURL: 'https://docslurp.io/v1', apiKey: process.env.DOCSLURP_API_KEY });
const completion = await client.chat.completions.create({
model: 'project:default', // the project to answer from
messages: [{ role: 'user', content: 'What changed in the 2026 travel policy?' }],
// @ts-expect-error DocSlurp extension, sent as part of the request body
docslurp: { tier: 'balanced' },
});OpenAI-compatible chat covers the rest of the Chat Completions contract.
Ingestion categories
Projects choose ingestion by category: quality (fastest, balanced, smart; the former fast, auto, good still work), visuals, embedding.profile, vision.profile and inferenceLane. Search settings choose the reranker as retrievalDefaults.reranker: keyword or model.
curl -X PATCH https://docslurp.io/v1/projects/default \
-H "Authorization: Bearer $DOCSLURP_SERVER_KEY" \
-H "Content-Type: application/json" \
-d '{
"pipeline": {
"quality": "smart",
"embedding": { "profile": "multilingual-local" },
"vision": { "profile": "good-fast-hosted" }
}
}'Settings apply to new uploads. To re-process existing documents, add "apply": "reindex"; that costs credits, so price it first by sending the same body to POST /v1/projects/:project/estimate.
Errors
A request that names a backend model, provider or llm:// URI is refused before any model call, with the categories to use instead:
{
"error": {
"code": "model_not_allowed",
"message": "Chat models are managed by DocSlurp. Send tier (fastest, balanced, smart) instead of a model; omit it for the project default."
}
}On Chat Completions the error's param is docslurp.model. A project or upload pipeline that names a model (for example pipeline.vision.modelUri) gets the same code with the field paths. Project and settings responses leave these fields out.
Glossary
Open the explainer →