Private beta — request access
A knowledge twin answers like its expert — drawing on their career, their writing and the way they make trade-offs. And it says when it doesn’t know.
Invited already? Create your account
Watch a twin think
Visitor asksFabric or Databricks for our lakehouse?
1Its own sources
2Knowledge graph & judgment
Expert’s ruleReversibility before hype — pilot one domain before you commit.
3Grounded answer
Public knowledge only · members-only sources stay private
Illustrative trace · every answer is assembled from the expert’s own sources, graph, rules and persona
How it thinks
Every message runs through eight stages before a word is written. And three loops keep running when nobody is asking — so the twin stays true to its expert over time.
Every answer · 8 stages
01Route — Intent & language
A question, a deliverable request or a URL to audit each go to the right service — automatically. The answer follows the visitor’s language.
On every new piece of knowledge
Each new statement is compared with what the twin already believes. A duplicate corroborates it, a refinement extends it, a contradiction is logged for the expert — and the expert’s own interview takes precedence. What it learns from conversations waits for approval.
Contradiction logged · the interview supersedes
Every night, only what changed
Confidence decays with a 90-day half-life unless knowledge is re-validated, and stale facts are flagged for review. Duplicate entities merge, orphan concepts get linked, thematic communities are rebuilt.
Continuously, against the real expert
A strategist picks what to ask next — failed questions, stale knowledge, uncovered domains, weak benchmark answers. A challenging interview digs in, the answers are distilled into graph, beliefs and persona, and a judge re-scores the twin on four axes.
judged against the expert’s own reference answers next interview topics
Expert network
The twin consults, behind the scenes, the twin whose knowledge actually covers the question — and brings its answer back, credited. The visitor never leaves the conversation. And if nobody knows yet, the question reaches the right expert.
Claire’s twin
Cybersecurity · the visitor’s conversation
Which architecture would you use for a multi-agent RAG in production?
Behind the scenes
Waiting…
Asks the one who knows
Among the twins of that field, the one whose knowledge actually covers the question.
Credits the answer
Brought back as a quote, attributed to that twin, with its own confidence.
Interrupts no one
It happens behind the scenes: no one’s conversation is touched, and the asking twin pays.
Gaps reach the right expert
If that twin doesn’t know yet, the question joins its expert’s next interview.
Only public twins whose expert agreed to be consulted take part — and only with their public knowledge.
For experts
A chatbot predicts plausible text. A knowledge twin reasons from an expert’s own material — and can prove where every answer comes from.
Résumé, LinkedIn, GitHub, GitLab, Confluence, Teams, SharePoint and documents become concepts and relations — each one traceable to where it came from.
Kafka←GitHub repo · Confluence pageevery concept keeps the source it came from
Public or members-only, per source. What it learns from conversations waits for your approval. Secrets are stripped first.
It produces real PowerPoint, Word and Excel files, embeds on your own website, and consults the expert twin, behind the scenes, when a question is outside its field.
It answers in English or French, in the register you describe — concise and factual if that is how you speak.
A chatbot
A knowledge twin
ChatbotAnswers from the internet at large
TwinAnswers from the expert’s own career, documents and code
ChatbotSounds confident even when it is wrong
TwinShows its confidence — and abstains when it doesn’t know
ChatbotSame generic voice for everyone
TwinApplies the expert’s trade-offs, measured against their real answers
ChatbotLearns nothing, or learns anything
TwinLearns from use — only what the expert approves
ChatbotAnswers everything, even outside its depth
TwinOutside its field, consults the twin who knows — and credits it
ChatbotA chat window
TwinA colleague: deliverables, your website, other twins
How it works
Résumé, LinkedIn, GitHub, GitLab, Confluence, Teams, SharePoint: a first draft of your twin is written from your own sources — nothing is made up.
A challenging interview captures how you make trade-offs; a benchmark measures whether it truly sounds like you.
Your own address and an embed for your site: your knowledge answers — in English or French — even when you’re away.
Content from your connected tools stays members-only until you publish it, and you approve what your twin learns. Secrets are stripped before any processing.