FactRule turns your databases, documents, and code into a living knowledge layer — a company memory that never walks out the door — that your whole organization can question in plain English. Every number is checked against your data before anyone sees it, or honestly refused. It runs inside your network; nothing reaches us.
A reasoning brain, hands to act with — and a memory of how YOUR business works. The first two are commodities. The third is your moat, and it's the piece that's been missing.
OpenAI, Anthropic, Azure, or a model running entirely inside your network. Swappable at any time — the intelligence is rented, your knowledge isn't.
A living graph of your databases, documents, and code — plus every settled meaning, every verified fact, every human decision. It grows with every question and it belongs to you.
The systems you already run become deterministic tools your agents use — ticketing, pricing APIs, MCP servers — each action permission-checked and recorded.
Once FactRule holds your knowledge layer, it can be whatever expert your organization is missing — and it gets sharper every day it works for you.
Succession planning, solved differently: your veterans teach FactRule once — every answer they confirm, every meaning they settle — and the company keeps that knowledge forever.
| FactRule | Generic AI chatbot on your DB | BI dashboards | Cloud-warehouse AI | |
|---|---|---|---|---|
| Answers verified against live data before you see them | ✓ | ✗ guesses | ✓ but pre-built only | ✗ generates, rarely verifies |
| Refuses honestly instead of picking a meaning silently | ✓ | ✗ | n/a | ✗ |
| Learns your business language durably (decide once) | ✓ | ✗ per-chat | ✗ hardcoded | partial |
| Ad-hoc questions, no build queue | ✓ | ✓ | ✗ backlog | ✓ |
| Runs entirely inside your network | ✓ | rarely | varies | ✗ their cloud |
| Governed agents with audit + approvals | ✓ | ✗ | n/a | partial |
| Works in Slack + your AI tools (MCP) | ✓ | varies | ✗ | partial |
We publish our prices (including enterprise), our security architecture and its limits, and a live demo you can stress-test without talking to anyone. Every claim on this site is checkable — the egress audit, the verification receipts, the refusals. We're recruiting founding customers: hands-on engineering support, roadmap influence, and pricing locked for two years. Talk to a founder.
A live FactRule instance holding the knowledge layer of a sample portfolio-services company. Ask it something a chatbot would guess at — and watch it prove the answer instead. No signup, nothing to install.
Open the live demoFactRule is one standard MCP connection. Point Cursor, Claude, or your own agent platform at it and every AI tool you run gains the company knowledge layer — with the verification certificate attached to every answer.
No — and structurally, not as a promise. Your data never reaches FactRule the company. What the product "learns" (verified facts, settled meanings, examples) is stored on your own volume, in documented formats, and is used only to answer your own organization. If you choose a cloud LLM, prompts go to YOUR provider account under your terms; fully in-network models are supported and recommended for regulated data.
Numbers pass live checks against your database before display — values exist, joins hold, an independent path agrees — and arrive with the receipt. When FactRule cannot verify, it refuses and says exactly why. Ambiguity becomes a human decision, once, on the record.
LLMs: OpenAI, Anthropic, Azure OpenAI, or any in-network OpenAI-compatible server (vLLM, Ollama) — swappable anytime. Data: SQL Server and PostgreSQL families plus SQLite today, alongside documents (local, SMB, Confluence, SharePoint) and source code. The roadmap is driven by founding customers.
OIDC single sign-on against your IdP (Okta, Entra, Ping), Windows-integrated auth via proxy mode, role separation (admins, adjudicators who may settle meanings, everyone else asks), and a tamper-evident audit log. SAML-only IdPs bridge at the IdP layer today — we say so rather than pretend.
One Docker container on any server in your network; a read-only database login is all it needs. Most teams see their schema mapped within the hour and their first verified answer in about 30 minutes.
Paid licenses fail open: serving never stops — new learning pauses until renewal. There is no remote kill switch, deliberately. Everything FactRule learned lives in your network in documented formats, and one command exports it. The exit is honest because we designed it that way.
Yes — the Sovereign edition activates with an offline license file, embeddings are baked into the image, and the built-in egress audit proves zero external connections to your security team.
Nothing, for 30 days, full product, no card. After that it stops answering (your data stays exportable) — there is no free tier, so nobody builds on something we quietly limit.