Skip to content
wirebase
Product · Knowledge

Answers from your own data.

Upload PDFs, docs, and websites. Ask questions. Get answers grounded in your sources with inline citations.

Knowledge base with uploaded documents, chunk counts, and semantic search
Cited answers, never hallucinated

“Per the Q3 OKR review (page 12), the marketing team committed to 200 net-new SQLs by end of quarter.” — Wirebase, citing your own document.

Upload anything

Drop in your documents.

Upload PDFs, Word documents, Markdown files, plain text, and web pages — Wirebase chunks and embeds them automatically using your chosen embedding provider. Knowledge bases are scoped per organization so each team's data stays separate. Re-index on update so the model always answers from the latest version of your docs.

  • PDF, Word, Markdown, plain text, websites
  • Automatic chunking and embedding
  • Per-organization, per-team scoping
  • Continuous updates — re-index on change
Document upload
Cited answers

Trust, but verify.

Every answer includes inline citation markers that point back to the exact chunks the model used. Click any marker to expand the source text, then jump to the original document to read it in full. When the model's answer doesn't match the source, you'll know immediately — citations make hallucinations visible.

  • Inline citation markers in every answer
  • Click to expand the source chunk
  • Jump to the original document
  • Confidence scores per citation
Citations in answer
Built on pgvector

Your data, your database.

Vector search runs inside your existing PostgreSQL via pgvector — there's no third-party vector database to provision, pay for, or worry about. Bring your own embedding provider (OpenAI, a local model, or a custom endpoint), and data isolation is enforced at the row level so one organization can never retrieve another's documents.

  • pgvector inside your existing PostgreSQL
  • Configurable embedding provider (OpenAI, local, custom)
  • Re-rank optional, deferred to v2
  • Per-org isolation enforced at the row level
pgvector architecture
What's inside

Retrieval that holds up.

Source attach

Files, URLs, raw text — all chunked the same way.

Semantic search

pgvector cosine similarity with metadata filters.

Org isolation

Knowledge never crosses organization boundaries.

Re-index on change

Updates flow through automatically.

Cited answers

Inline markers tied to source chunks.

Embedding choice

Bring your own embedding provider.

Get your team on one AI platform.