Skip to content

Private AI Knowledge

Your private AI: trained on your documents, connected to your work.

Upload SOPs, manuals, and reference docs, and connect the records your team already tracks. Ask anything in plain English — get answers with citations, and let it draft notes or create tasks with your approval.

How indexing works

When you upload a document, we OCR scanned pages, chunk the content into passages, and embed each chunk into a private vector index. When the source changes, we re-index automatically so answers stay current.

  • PDFs and scanned PDFs
  • Word, plain text, images with text
  • Per-tenant embedding store
  • Re-indexing on updates

Citations on every answer

Every response includes the exact source paragraph and page number. You can hover for surrounding context or click through to open the original document — no floating claims.

  • Inline citation chips
  • Source-paragraph preview panel
  • Hover to see surrounding context
  • Click to open the original document

It works from your records — and acts

Beyond documents, the assistant reads the clients, policies, renewals, tasks, and notes your team already tracks — then takes the next step. It drafts a note or creates a task and proposes the change for a one-click confirm, so nothing is written without a person's say-so.

  • Reads clients, policies, renewals, and tasks
  • Drafts notes and creates tasks for you
  • Every write waits for your approval
  • Cites the record or paragraph it used

Conversation history

Threads stick around. Each user owns their conversations and can search across past Q&A, resume an old thread, or share it with the team when something is worth keeping.

  • Per-user conversation log
  • Resumable threads
  • Search across past Q&A
  • Optional sharing inside the team

Per-tenant isolation

Documents, embeddings, and conversations are strictly separated by tenant. Nothing crosses between customers, and your data is never used to train shared models.

  • Isolated storage bucket
  • Isolated vector index
  • Isolated chat history
  • Never used to train shared models

Supported file types

Most of what teams already store as reference material works out of the box. If a file has text we can read, we can index it.

  • PDF (text and scanned)
  • Microsoft Word (.docx)
  • Plain text and Markdown
  • Images with text (PNG, JPG)
  • Common spreadsheet exports

Limits & latency

Answers stream in as the assistant works, so your team sees progress immediately. Per-tenant rate limits and a monthly token budget keep costs predictable and prevent the long-tail timeouts that show up under load.

  • Streaming responses
  • Per-tenant rate limits
  • Per-tenant monthly token budget
  • No long-tail timeout failures

Built so your most sensitive documents can live here.

Tenant isolation

Your documents, conversations, and embeddings live in a workspace only your team can reach.

Encryption in transit and at rest

TLS 1.3 in transit, AES-256 at rest, on isolated cloud storage.

Citations on every answer

No floating claims. Every response links to the exact source paragraph.

You approve every write

The assistant proposes actions — a task, a note — and waits for a human to confirm before anything changes.

Never used to train shared models

Your data is never used to train models that touch any other customer.

Compliance-ready

SOC 2 Type II in progress. DPA on request.

Stop searching folders. Put your AI to work.

A 20-minute walkthrough on your own documents and your own book. No slideware, no scripted demo — your real questions, your real answers, and the follow-up work it can tee up for you.