Agentic RAG over your own documents

Answers from your documents, with the receipts

Helpdesk turns the files your team already has into a knowledge base you can ask questions. Every answer is grounded in your own content and linked back to the passage it came from.

Get startedSee how it works

Self-hosted. Your documents and embeddings stay in your own Postgres.

Product Docs· 12 documents indexed
What’s our refund window for annual plans?
Annual plans can be refunded in full within 30 days of the renewal date. After that the plan stays active until the end of the term and no partial refund is issued.
Sourcesbilling-policy.pdfp. 4terms-of-service.md§ 7.2
Retrieval traceRetrieved 8 chunksGraded 5 as relevantAnswered from source

Built for answers you can actually rely on

A general-purpose assistant guesses. Helpdesk retrieves, checks what it retrieved, and shows its work.

Answers you can check

Responses are generated only from the passages retrieved out of your own documents, and every answer carries the sources it was built from.

Retrieval that corrects itself

The agent grades what it retrieved before answering. If the context is weak it rewrites the question and searches again rather than guessing.

Bring your own documents

Drop in the files your team already keeps. They are chunked, embedded, and made searchable in the background — no manual tagging.

Inspect the retrieval

Open any question and see the exact chunks that were pulled and how they scored, so you can tell a bad answer from a bad document.

Scoped by project

Keep separate bodies of knowledge apart. Each project has its own documents, its own index, and its own chat threads.

Streams as it answers

Answers arrive token by token. The agent's internal grading and rewriting steps stay behind the scenes, so you only see the response.

From a folder of files to a working knowledge base

Three steps, and only the first one is yours to do.

  1. Step 1

    Upload

    Add documents to a project. Files go straight to your own object storage, and processing is queued the moment the upload lands.

  2. Step 2

    Index

    A worker extracts the text, splits it into passages, and embeds each one into a vector index sitting next to your data in Postgres.

  3. Step 3

    Ask

    Ask in plain language. The agent searches the index, keeps only the passages that actually answer the question, and cites them.

Retrieval quality is measured, not asserted

Changes to the retrieval pipeline are checked against a benchmark with significance testing before they ship, so improvements are real ones rather than noise.

Evaluated on a public question-answering benchmark

Built on
  • PostgreSQL + pgvector
  • LangGraph
  • Gemini embeddings
  • NestJS
  • Next.js

Stop searching. Start asking.

Create a project, upload your first document, and ask it something.

Get started