Wikis describe the work. VectorDBs show your agents what is similar. Seyn observes why work actually happens the way it does, and gives your agents what is useful.Seyn is a knowledge extraction engine. It connects to the systems where your organisationâs work already lives (SharePoint, Teams/Slack, email, document archives, operational tools), ingests the raw activity, and extracts the processes, rules, and patterns that actually run the business. The output is a versioned knowledge library where every claim traces back, hop by hop, to the raw source records it was derived from.
These docs are alpha. Seyn is in active development. Some capabilities described here are still in testing, and every page flags them explicitly. The current state of each feature lives on Status.
The problem
How an organisation really operates is almost never written down. It lives in peopleâs heads, in ten thousand Teams/Slack messages and emails, in the folder structure of a shared drive, in the gap between the official process diagram and what the team actually does on Thursday afternoons. Search doesnât fix this. Similarity search finds documents that look like your question; it canât tell you who approves what, in which order, or why the process forked last quarter. Useful is not the same as similar. Seynâs answer is extraction, not search:- Interviews ask people what they think they do. Seyn watches what the data says they do.
- A process map is an opinion. A process rule with provenance is a claim with receipts.
- Similar is a guess. Extracted, reviewed, versioned knowledge is an answer.
The principle
Seyn is built on three primitives, and everything else composes from them:| Primitive | What it is | Read more |
|---|---|---|
| Events | Raw activity normalized into one schema: who did what, to which entity, when. | Events |
| Knowledge | What the events mean: processes, rules, and patterns, extracted and versioned. | Knowledge |
| Connectors | How the activity gets in: read-only ingestion from any system, through prebuilt connectors or your own. | Connectors |
What you can build on it
Grounded agents
Give any MCP-capable agent org-scoped access to extracted process knowledge. Your agents stop guessing how your business works.
Auditable answers
Every rule traces four hops back to source records. Built for teams where âthe model said soâ is not an acceptable answer.
Conversational knowledge
Ask questions in plain language, get cited answers assembled from hybrid retrieval over versioned, human-reviewed knowledge.
Typed integrations
A read-only v1 API with TypeScript and Python SDKs for compliance tooling, internal apps, and BI pipelines.
Performance
Numbers we hold ourselves to on real client corpora:- Sub-second hybrid retrieval. Structured, full-text, and semantic signals run concurrently inside one PostgreSQL instance; queries return well under a second at production corpus sizes.
- Single-call provenance. The full four-hop audit chain for any rule resolves in one request.
- Incremental everything. Delta sync and content-hash deduplication mean re-ingesting an unchanged corpus costs almost nothing, and analysis cost scales with new data, not corpus size.
Where to start
Get the mental model
Read Core Concepts for the vocabulary, then Architecture for how ingestion and querying fit together. Fifteen minutes, and every other page will make sense.
Make a real query
The Quickstart takes you from an API key to a natural-language query against extracted knowledge in under ten minutes.
Check what's live
Status tells you whatâs in testing and whatâs coming, with reasons.
Stability
The/v1 API is the stable public contract. Breaking changes ship as /v2; additive changes ship within /v1, so write integrations to ignore unknown fields. Features marked In testing carry no compatibility promise yet.
Seyn is currently in private alpha. To evaluate it on your own data, request access.