Use case
The knowledge base API for grounded answers.
Add retrieval-augmented, cited answers to your product with the OAuth CLI or one JSON endpoint—compact citations by default and streaming when you opt in.
Build versus operate
Ship the feature, not the retrieval pipeline
Building retrieval-augmented answers in-house means owning parsers for a dozen file formats, an embedding pipeline, a vector store, passage-splitting heuristics, retrieval tuning, citation plumbing, and the infrastructure to keep it all fresh. That is a platform project, and it starts before the first useful answer.
kbDrop collapses it to two steps: create a knowledge base from documents or a website, then call one endpoint with a scoped key. Your code renders events; kbDrop operates the pipeline.
What the endpoint gives you
Grounded answers as a contract
Retrieval and citations included
Each request runs retrieval and returns one authoritative answer with compact, permission-safe citations that your product or agent can follow.
Keys scoped to what you plug in
API keys are stored as hashes, shown in full only at creation, rate limited, independently revocable, and unable to query any knowledge base not plugged into them.
JSON first, streaming on demand
Ordinary requests return a complete JSON result. Opt into the existing response.started, citation, response.delta, response.completed, and error stream when incremental UI matters.
One request
POST a question, receive completed JSON
Send a scoped key as a Bearer token and omit the response selector for one response.completed JSON object, answered from every knowledge base plugged into that key. The quickstart also covers OAuth CLI, controls, structured output, and explicit SSE.
curl --fail-with-body --request POST \
--url 'https://kbdrop.io/v1/messages' \
--header 'Authorization: Bearer kb_live_...' \
--header 'Content-Type: application/json' \
--data '{"input":"How do I reset the device?","citation_detail":"compact"}'What teams build with it
- In-product help that answers from your docs site with citations your users can open.
- A support assistant grounded in the manuals and macros your team already maintains.
- Internal tools that answer from a ZIP of policies, contracts, or runbooks.
- Prototypes that need credible retrieval this week, not after a quarter of pipeline work.
Questions, answered
Knowledge base API FAQ
Is this a RAG API?
Yes. kbDrop is retrieval-augmented generation as a managed service: source processing, indexing, retrieval, answer generation, and citations run server-side, and your application receives one compact JSON completion by default.
How are API keys secured?
Keys are scoped to a single knowledge base, stored as hashes, shown in full only when created, rate limited, and revocable at any time from the knowledge base's Developers tab. Keep them on your server and send them as Bearer tokens.
What does the API cost?
API answers draw from the same monthly cited-answer quota as chat. The Free plan is enough to prototype against, and the Starter and Scale plans raise the quotas—see the pricing page for the current numbers.
Are there SDKs?
Use the official @kbdrop/cli for OAuth, retries, and stable machine output, or call the plain JSON API from any HTTP client. Server-sent events remain an explicit option.
Can an AI agent integrate kbDrop for me?
Yes. kbDrop publishes an open Agent Skill for Codex, Claude Code, Gemini CLI, Cursor, and other compatible harnesses. Install it with npx skills add https://kbdrop.io --global, then let the agent use the scoped key from your secret store.
How do I handle rate limits?
Rate-limited and retryable responses include Retry-After. Preserve one idempotency key, honor that delay, and use bounded retries; the official CLI does this automatically.
First cited answer in an afternoon
Create a knowledge base, authenticate the CLI, and return a compact grounded answer before the day ends.
Create a free account