The company knowledge base
ourknowledge.dadam.appTeams sign up, invite the whole company, and ask in any language. Every answer is assembled from your own documents — and footnoted back to the paragraph it came from.
Invite by link · Postgres + pgvector · English & 한국어
What’s our refund policy for annual plans?
Annual plans are fully refundable within 30 days of purchase.[1] After day 30, the unused months become account credit rather than a cash refund.[2]
Refunds return to the original payment method and settle within five business days.[1]
GapNothing in the retrieved sources covers mid-term downgrades — worth confirming with Finance.
Billing Policy 2026.md
Refunds › Annual plans
help.acme.com/billing/refunds
Credits › Processing times
Architecture
Cerebras published how they built the internal knowledge base their engineers query thousands of times a day; OurKnowledge follows that architecture — one embeddings table, hybrid retrieval, synthesis that cites — rebuilt as a multi-tenant product for your company.
Pasted text, uploaded Markdown and fetched URLs all normalize into one Postgres table with pgvector. One row schema, one query path, one place to index — adding a source type later changes nothing downstream.
chunks(org_id, heading, content, embedding, tsv)
Full-text search catches the exact token; embeddings catch the paraphrase. Both legs are ranked separately, fused with reciprocal rank fusion at k=60, then damped by an exponential age-decay term so stale docs quietly lose.
score = Σ 1/(60 + rank) × exp(−age)
The model only sees numbered blocks pulled from your workspace, and must mark each claim with the block it came from. When the evidence is thin or contradictory it is required to say so rather than fill the gap.
[n] → document · heading · passage
Ingest
paste · file · url
Chunk
heading-aware, ~1,200 chars
Embed
1,536-dim, batches of 64
Retrieve
top 40 per leg
FTS ts_rank_cd
VEC embedding <=> q
→ RRF k=60 · age decay
Answer
streamed, cited [n]
How it works
Three steps, and nothing to install or configure. The first two take a minute each; the third is where the work actually pays off.
Sign up with an email and a password, name your company, and you have a workspace. Everything inside it — documents, chunks, conversations — is scoped to that organization from the first request.
email + password · no install
Generate an invite link and drop it in Slack; anyone with the link joins your workspace after signing in. Roles decide the rest: owners and admins manage knowledge and members, members read and ask.
invite links · owner / admin / member · 7-day expiry
Paste a policy, upload Markdown or plain text, or point at a URL and we fetch and strip it down to readable text. It is chunked, embedded and searchable in seconds — then ask, and read the citations.
paste · files · urls · ask & search
What you get
Six things v1 does properly, instead of a dozen it does halfway.
Get started
Create your company, share one invite link, add the first document. You can be reading a cited answer in the next five minutes.