ChatGPT that knows the company. Privately.
A done-for-you retrieval-augmented system over a firm's own documents — contracts, policies, tickets, manuals — with citations and permission scoping. It answers the questions that used to need a partner, and never sends data to a public model.
In short
- What it is: A private ChatGPT over a company's own knowledge, with citations.
- Price: from $999 one-time, with an optional $199/mo care plan.
- Time to launch: installed, branded and wired to Stripe in about 48 hours.
- Best for: Legal, finance, healthcare, ops.
30-day launch guarantee · you own the code and the customers
stack: RAGFlow · Qdrant · Letta · OpenRouter
best for: Legal, finance, healthcare, ops
time to first sale: 3–6 weeks
margin: High
✓ installed, branded & wired in 48h
Buy once. Deploy for 1, 20, or unlimited clients.
The licence is one-time and sets how many businesses you may deploy for. The care plan (updates, monitoring, support) is optional — bring your own keys and skip the monthly entirely.
Single
1 client
one-time · you own the code
Studio
up to 20 clients
one-time · you own the code
Unlimited
unlimited clients
one-time · you own the code
Care plan from $199/mo, optional · or run it yourself with your own model keys.
Watch the Ask-Your-Docs RAG demo.
A short walkthrough of the dashboard, the automations and the client-facing experience. Enter your name and email to unlock it.
- Dashboard & client portal tour
- The automations running live
- What the buyer actually receives
Why this sells.
Data stays put
Self-hosted or in the buyer's cloud, with their keys. The single objection that kills AI deals, removed.
Citations build trust
Every answer links back to the exact clause and page. Professionals verify in seconds instead of re-reading files.
Retention is structural
When the team relies on it to work, cancelling means losing the answers. That's a very long subscription.
Entry price
Care plan
Time to first sale
Features.
Any document type
PDFs, Word, spreadsheets, email archives, tickets, wikis and scanned images (OCR).
Citations & sources
Every answer points to document, page and clause. No unsourced claims.
Permission scoping
People see only what they're allowed to. Roles filter retrieval before it happens.
Any model, any host
OpenRouter for hosted models or a local open-weight model for total privacy.
Slack, web & API
Ask from Slack, the browser or an API — wherever the work already happens.
Evals & monitoring
Retrieval quality scored continuously, with alerts when accuracy slips.
Live in 48 hours.
Point at the corpus
Share the drives, wikis and mailboxes to ingest.
We build the pipeline
Chunking, embeddings, reranking and citation logic.
Permissions & evals
Access roles wired, retrieval accuracy tested.
Deploy where they want
Their cloud or on-prem, with their model keys.
Included automations.
- ingest_documents.py — OCR, chunk, embed and upsert at scale.
- eval_retrieval.py — Score hit-rate and faithfulness on a golden set.
- query_cli.py — Ask the corpus from the terminal for demos.
- permission_sync.py — Mirror access roles into retrieval filters.
What you pay us — what you charge them.
You pay us
one-time install, self-hosted or your cloud
- Ingest pipeline + evals
- Citations & permissions
- Slack / web / API
- Care plan optional +$199/mo
You charge them
setup, plus $499–$2,500/mo retainer
- Scoped to corpus size
- Private deployment premium
- Annual contracts are normal
- You keep 100%
The maths
vs a support agent on monthly revenue, at half the churn
- Regulated industries can't switch off the source of truth
- Compliance budget, not IT budget
- Long contracts are the norm
“We replaced a partner's three-hour research task with a cited answer in four seconds. Nobody wants to go back.”
Legal operations · 300+ seat deployment · on-premise
FAQ
How big a corpus can it handle?
From a few hundred documents to millions of pages. Vector databases scale horizontally, and we size the deploy to the corpus.
Can it run fully offline?
Yes — with a local open-weight model and a self-hosted vector store, nothing leaves the building.
How do you stop hallucination?
Grounded-only prompting, citations on every claim, and an abstain threshold. If the answer isn't in the corpus, it says so.
Who maintains the index?
An incremental sync keeps it current. Add it to the care plan and we watch quality for you.
Own your AI business this week.
$999 to install. Sell it to one client and you're in profit.