Know-how,
kept in-house.
KnowHQ turns the documents your company runs on into knowledge bases, semantic search and AI agents, then runs them where your data is allowed to live.
QuestionHow long do we keep supplier invoices?
Accounting documents are retained for ten years after the financial year ends.
records-policy.pdf, section 4.2
What you know is scattered. Fixing that shouldn’t mean giving it away.
Policies sit in shared drives, answers in old email threads, procedures in scanned PDFs. AI could reach all of it, but legal, clients or regulators often say no to public tools. We start from that constraint and build within it.
Knowledge
Your documents become one source people and software can query, with every answer traced to its page.
AI
Claude, Codex or open-weight models, chosen per job. You can change them later without rebuilding.
Privacy
Where the data lives and what leaves your network are decided first and written down.
Four services, one body of knowledge.
Start with one. Each builds on the same indexed documents, so the next costs less to add.
- KB-01
Knowledge base creation
We turn PDFs, office files, scans, websites and repositories into one organized source your people and software can query.
- Connectors for file upload, websites, S3 buckets and Git
- Scanned pages digitized by vision models, with structured fields where you need them
- Separate projects, so each team sees only its own material
- SR-02
Semantic search
People search with the words they have, and still reach the passage that answers them, even when it uses different words.
- Answers that link back to the document and section they came from
- Vector search and reranking tuned against your real questions
- Delivered as chat, a REST API or an embedded website widget
- AG-03
Agentic applications
We build agents with Claude, Codex and open-weight models that work from your knowledge base and act through tools you define.
- Workflow agents that fill forms, update records and open tickets
- Coding agents set up for your repositories, conventions and review process
- Human approval on anything that is hard to undo
- OP-04
On-prem and local AI
We install and tune private AI on servers you own, for work that cannot go to a public API.
- Hardware sizing and open-weight model selection for your workload
- Models, document store and vector index inside your network
- Sign-in through your identity provider, using OpenID Connect
One foundation under every project.
We don’t rebuild retrieval from scratch for each client. Every engagement starts on an open-source knowledge platform that we install, extend and operate.
- 01
Import
File upload, websites, S3 buckets and Git repositories.
- 02
Read
PDF, Word, Excel and PowerPoint. Scans go through vision models page by page.
- 03
Split
Documents become overlapping sections, so search finds the passage and not only the file.
- 04
Index
Sections are embedded and stored in PostgreSQL with pgvector.
- 05
Retrieve
Semantic search and reranking bring back the strongest evidence.
- 06
Deliver
Chat, REST API, batch document review and a website widget that can act on the page.
Model choice stays open
Connect Anthropic, OpenAI, Google Gemini, Mistral, Cohere or Groq, or run open-weight models on your own hardware.
No lock-in
The platform is open source and self-hostable. If we ever part ways, your system keeps running.
Decide where it runs before anything gets built.
Three ways to deploy. Each one says plainly what leaves your network.
| Deployment | Where models run | Documents and index | What leaves your network |
|---|---|---|---|
| Your hardware | Open-weight models on your servers | Your servers, inside your network | No documents, questions or answers |
| Your private cloud | Open-weight models on your instances, or your cloud’s model service | Your cloud account, in the region you choose | Nothing beyond the cloud account and services you approve |
| Hosted models, your accounts | Claude, OpenAI and others, under your own API keys and contracts | Your environment | The question and the passages retrieved for it, sent only to the provider you chose |
We don’t train on your documents
Your material is used to serve your system and nothing else.
Your keys, your contracts
When a hosted model makes sense, the provider relationship is yours.
Access follows your rules
Sign-in goes through your identity provider, and projects keep teams’ data apart.
How an engagement runs.
Map
We go through your sources, who may see what, and what must never leave. You get a written decision on where the system runs.
Prove
A pilot on real documents, judged on the questions your team actually asks, with sources you can check.
Install
Production deployment in the agreed environment, with sign-in, permissions and agents connected to your tools.
Hand over
Documentation and training for your team. Run it yourselves, or have us operate it for you.
Where this earns its keep.
Contract review
The same checklist across a batch of contracts, returned as a matrix with flags for what needs a lawyer.
Scanned records
Certificates, manifests and survey reports turned into structured data without retyping them.
Internal document hub
Policies, reports and procedures, where a question returns the passage that answers it.
Customer support
Answers drawn from your real help content, with a clean handoff when the answer is not there.
Engineering teams
Claude and Codex configured for your codebase, with shared skills and review rules.
Regulated work
Useful AI for teams whose clients, auditors or regulators rule out public tools.
Bring us one messy folder.
Tell us what documents you have and which questions people keep asking. We’ll say what it takes to make them answerable, and where the system should run.