Artificial intelligence · Search API for RAG
Give your AI assistants the answers already in your documents.
Upload articles, manuals and policies as text or Markdown and retrieve the right passages for every question. Hybrid search, optional reranking and relevance feedback, with no vector database, embedding models or infrastructure to set up.
curl -X POST 'https://api.respawntech.dev/v1/tools/rag/collections/3c9e7a52-1b4d-4f0e-9a6c-2d8e5f7a1b3c/query' \
-H "X-API-Token: $RESPAWN_API_TOKEN" \
-H 'Content-Type: application/json' \
-d '{ "query": "como mudo minha senha?", "top_k": 3 }'Use cases
Semantic search over your documents for AI chatbots, assistants and agents.
Support chatbot
Answer customers from your help center, citing the right article.
Internal assistant
Let your team ask about processes, policies and technical documentation.
AI agents with context
Connect the knowledge base to agents through the MCP server and let the agent search on its own.
How it works
From sign-up to first call in minutes
Create a knowledge base
One call creates the knowledge base in the language of your documents.
Upload your documents
Text or Markdown; indexing is asynchronous and you can track its status.
Search and answer
Send the question, get the most relevant passages and use them as context for your model.
Features
Built to run in production
Hybrid search
Combines meaning (embeddings) and keywords, so it finds both a paraphrased question and an exact error code.
Rerank when you need it
Accurate mode reorders results with a dedicated model, and standalone rerank sorts any list of texts.
Feedback that improves search
Confirm or reject passages: similar questions then surface the confirmed ones first.
Zero-downtime updates
Resend a document with the same identifier and the old version keeps answering until the new one is ready.
Metadata filters
Tag documents with a category, product or language and filter on them at search time.
Multilingual
A multilingual embedding model plus search dictionaries for Portuguese, English, Spanish and French.
Pricing
Pay for usage, not a subscription
Searches and reranking include a free quota every day; after that, they are billed in blocks. Indexing is billed per chunk, and storage is free up to 10,000 chunks.
Knowledge base — per use
| Item | Credits |
|---|---|
| Ingestion, per indexed chunk | 0.015 |
| Search | 0.25 |
| Search served from cache | 0.02 |
Rerank in accurate modemode: accurate | +0.08 |
| Standalone rerank, per call | 0.08 |
Knowledge base — free every day
After the free quota, calls are billed in blocks of 100.
| Operation | Free per day |
|---|---|
| Searches | 100 |
| Searches served from cache | 100 |
| Rerank in accurate mode | 20 |
| Standalone rerank | 20 |
Knowledge base — storage
| Item | Credits |
|---|---|
| First 10,000 chunks | Free |
| Each block of 10,000 chunks | 0.1 / per day |
FAQ
Frequently asked questions
Do I need a vector database or an embedding model?
No. The API handles chunking, embeddings, the index and reranking. You send text and run searches.
Which formats can I upload?
Plain text or Markdown, sent as JSON, up to 5 MB per document. For PDF or Word files, extract the text before uploading.
Does the API generate the final answer for the user?
No. It returns the most relevant passages, and you pass them to the language model of your choice to write the answer.
Can I use it with AI agents?
Yes. The same knowledge bases are available through the MCP server, with a spending limit per key.
Is it available to every account?
The tool is in beta and rolling out gradually. Get in touch to join.
Start integrating today
Create your account, generate a token and make your first call in a few minutes.