Ts-Knowledge
Sign inCreate an account

RAG · Structured knowledge · Knowledge Snowflake · MCP

Your knowledge — found, understood, and plugged into your AI.

Ts-Knowledge indexes your documents so your AI can find the right passage, extracts the people, facts, relationships and events they contain — each with its evidence — and makes it all available to Claude, ChatGPT, Gemini or Mistral through an MCP server.

Create an accountSee how it works

No LLM lock-in. Your corpus stays yours — and so do your model keys.

Screens shown with the French interface.

Ts-Knowledge dashboard: spaces, documents, indexed chunks and recent activity.

Three layers, one corpus

This isn't yet another chatbot. Ts-Knowledge does the work your AI can't do on its own: know what your documents contain, prove it, and serve it up under the permissions you've set. Writing the answer is left to the AI you've already chosen.

  1. Your documentsPDF, DOCX, Markdown, websites, Google Drive, GitHub
  2. FindHybrid search, cited passages, and a “not enough evidence” verdict
    UnderstandEntities, facts, relationships, events — each backed by evidence, arranged into a Snowflake
  3. ConnectMCP server and REST API, under each key's permissions
  4. Your AIClaude, ChatGPT, Gemini, Mistral, an IDE agent, your own code

1 · Find

RAG that cites its sources — and knows when to say “I don't know”

The foundation of the product: retrieval-augmented generation (RAG) that finds the right passage — and can always tell you where it came from.

Walled off by space

Organization, project, case file, corpus: each space has its own documents and permissions. An API key only sees what you've opened to it.

Hybrid search

Meaning and exact wording, fused: a plain-language question and an article number both find their passage.

Structured citations

Title, chapter, page, article, source URL: every excerpt arrives with what you need to verify it — not just a blob of concatenated text.

A score that means something

Relevance is measured, not inferred from a rank: it compares from one search to the next, and can be thresholded on the client side.

An engine that can say “I don't know”

When the corpus doesn't hold the answer, it says so instead of returning its nearest neighbors — which is what keeps your agent from making things up.

Your formats

PDF, DOCX, Markdown, HTML, plain text — split along chapters, articles and paragraphs, never in the middle of a sentence.

See exactly what your AI will receive

The Playground shows raw retrieval: the passages selected, their relevance, their citation — chapter, section, page. Before you connect an agent, you already know whether your corpus has the answers.

  • Three modes — semantic, lexical, hybrid — to understand what brings a passage to the top.
  • An explicit verdict: the engine tells you when what it found isn't enough to answer.
  • The API payload copies in one click: what you test is exactly what your integration will send.
Ts-Knowledge Playground: a question, and three excerpts with their score and citation.

2 · Understand

Beyond RAG: structured knowledge

A search returns passages. Some questions call for more: which characters know the secret?, in what order did the events of Book 2 happen?, where do two chapters contradict each other? Ts-Knowledge reads your documents and draws out what they assert — entities, facts, relationships, events — keeping, for each one, the passage that proves it. It's a second layer, derived from the corpus, that adds to search without ever replacing it.

One profile per entity, sourced line by line

A character, a place, a department, a party to a case: the profile gathers everything your documents say about it, organized by theme, with the number of passages supporting each value.

  • Aliases recognized: “Maëlle”, “the keeper's daughter” and “M. Kerlann” all refer to the same person.
  • First and last appearance, key relationships, and the events the entity takes part in.
  • Every value opens onto its evidence: the passage, the chapter, the page.
An entity profile: identity, description, aliases, sourced facts, and a flagged contradiction about age.

Entities

Characters, places, organizations, objects — or whatever your field calls them: the vocabulary is configured per space.

Relationships

Who is whose friend, mentor or rival, in a graph you can navigate. Every link opens onto the passages that establish it.

Timeline

Events, dated or ordered by the narrative, with their participants and the document where they take place.

Inconsistencies

Two values that can't both be true — same context, same period — are flagged and ranked by severity, never settled on your behalf.

Relationship graph between entities: explicit links as solid lines, inferred links as dotted lines.
The relationship graph: a solid line if it's written, a dotted one if it's inferred.
Event timeline with a calendar, participants and the source chapter.
The timeline — dated when the text allows it, otherwise in narrative order.

New · the Knowledge Snowflake

Your whole universe, on a single map

Two hundred characters, four books, a side story, world rules and secrets not everyone knows yet: past a certain size, no list is enough. The Knowledge Snowflake — the “Flocon” screen in the app — arranges all of a project's knowledge into facets, book by book, and shows you at a glance what's solid — and what's missing.

One arm per facet, one ring per book

Characters, the world, affiliations, world rules, forms, arcs, “who knows what”, timeline, glossary, author's choices: each facet becomes an arm of the Snowflake, each book a ring. The bigger the dot, the more that book has to say.

  • Gaps jump out at you: an empty facet stays empty, never guessed — you know exactly what to write, document or extract next.
  • The hub and its branches: the universe holds the identity of characters and places, while each book and each folder of drafts is a branch — so a contradiction between a chapter in progress and a published book shows up right away.
  • Side stories in their rightful place in the story, without ever counting as read.
The Snowflake of a saga: ten facets radiating from the universe, one ring per book, and the coverage of each facet book by book.

Links across the whole series, without a single spoiler

The Snowflake graph colors each link by facet: who loves whom, who belongs to which clan, who knows which secret. Zoom, pan, and group by facet, by type or by book.

  • “Read up to Book 2”: one click, and the Snowflake shows only what a reader who stopped there can know. No name revealed later, no link that gives away what comes next.
  • What's written, what's inferred: a solid line for a citation, dashes for an inference — you can tell the difference at a glance.
  • Your agent reads it too: get_knowledge_flake gives it the project's map — where to look, and what's missing — before its very first search.
The Snowflake's links grouped by facet: characters, world, affiliations, arcs, secrets, world rules.

For a saga…

Books, side stories, notes and drafts from the same universe: your series bible, kept up to date on its own with every chapter you import.

…and for a business

Business objects, processes, rules, decisions, systems: the same Snowflake maps out an online store or a department, stage by stage.

Zero spoilers, by design

The reading limit bounds everything: excerpts, entities, facts, links and names. When in doubt, the Snowflake stays silent.

Knowledge you can verify

Knowledge extracted by an AI is only worth something if you can tell where it came from. Four rules govern this layer, and none of them can be switched off.

  • Nothing without evidence. Every statement is tied to the document, page and sentence that establish it. When no document supports it anymore, it disappears.
  • Cited, inferred or generated — never mixed. An extracted statement is marked “explicit” only if its quote appears word for word in the text; otherwise it is filed with the inferences, along with its confidence.
  • Different isn't contradictory. A form taken in another scene, a status that changes in chapter 20, a nickname: none of these is an inconsistency. The alert only fires for two values that can't both be true — and neither one overwrites the other.
  • Your decisions stick. A statement you dismiss stays dismissed, even when a new extraction finds it again.
Structured knowledge overview: entities, facts, relationships, events, evidence, inconsistencies, and a breakdown by origin.

Read by your model…

Turn on extraction for a space: every imported document is read in the background by the model you've connected, with your key and on your account. Search never waits for extraction.

…or by your agent, via MCP

Your agent can do the reading itself: the server hands it the instructions, the space's vocabulary and the text, then re-checks everything it proposes — under exactly the same rules.

Off by default, enabled space by space. The vocabulary starts from a template — generic, book, software, legal, e-commerce — and adapts to your field.

3 · Connect

Built for MCP, not bolted on

The MCP server is Ts-Knowledge's primary interface. Everything the app offers, your agent finds as tools: search, read, explore the structured knowledge — and even enrich it. The model that writes the answer remains yours.

Your agent searches, verifies, then answers

  • One URL, one key: your agent presents its key as a bearer token — nothing else to configure.
  • One key per agent: each key sees only the spaces you open to it, with the permissions you check — search, read, ingestion.
  • Cited answers: every excerpt and every fact arrives with its structured source, not as concatenated text.
  • We never send anything to a model: your agent comes and fetches what it needs.

Who knows the lighthouse secret, and since when?

get_knowledge_flakeSaga des Marées · read up to Book 210 facets · “Who knows what”: 13 statements, 2 secrets
search_knowledge“lighthouse secret” · Saga des Marées3 excerpts · Book 2, ch. 12, p. 143 · Book 1, ch. 3, p. 21 · …
get_knowledge_entityMaëlle Kerlann245 facts · friend: Yann Le Goff · mentor: Rozenn · 1 contradiction

Maëlle has kept the secret ever since her father entrusted her with the keys to the lighthouseBook 1, ch. 3, p. 21. Yann guessed it from the jettyBook 2, ch. 12, p. 143. The books disagree on how old she was at the time (16 or 17).

report_used_chunks2 excerpts were used

Thirty-one tools, from search to enrichment

Get oriented

  • list_knowledge_spacesthe spaces open to the key
  • list_knowledge_collectionsthe collections in a space
  • list_knowledge_labelslabels, to narrow the scope
  • list_knowledge_documentsdocuments and their status

Search

  • search_knowledgethe relevant passages, with their citations
  • research_knowledgeand, when the corpus falls short, your connected sources

Read

  • get_knowledge_documenta document and its metadata
  • get_knowledge_chunkan excerpt and its immediate context
  • get_document_outlinethe outline of a long document

Understand

  • list_knowledge_entitiescharacters, places, services, articles…
  • get_knowledge_entitythe profile: facts, relationships, events
  • get_entity_relationswho is linked to whom, and how
  • get_knowledge_flakea project's map: its facets, book by book, and what's missing
  • get_entity_evidencethe passages that prove it
  • list_knowledge_conflictsthe project's inconsistencies, with values and evidence side by side
  • list_knowledge_extractionshow far the reading of each document has progressed
  • get_story_rhythma book's pacing, measured against its author's structure grid

Contribute

  • start_knowledge_extractionthe instructions and the text of a document to read
  • submit_knowledge_extractionwhat the agent drew from it, re-checked by the server
  • merge_knowledge_entitiesmerge two entities that are really one, at your request
  • promote_knowledge_entityshare an entity with the whole series or the whole company, at your request
  • update_knowledge_entitycorrect an entity's name, nicknames or description
  • set_knowledge_entity_statusarchive an erroneous entity, or restore it
  • set_knowledge_statement_statusdismiss a false fact, relationship or event, on your decision
  • resolve_knowledge_conflictresolve an inconsistency from your own explanation
  • start_visual_reference_readingthe author's sketches, to read their style and features
  • submit_visual_reference_readinga sketch's style, then enforced on every illustration
  • start_visual_descriptiona character's facts, to write their illustration prompt
  • submit_visual_descriptionone image prompt per form, stored separately on the profile
  • set_document_reading_ordera series' reading order, so nothing gives away what comes next
  • set_story_grida story-structure grid, entered from your notes
  • start_structure_reviewa book's grid and chapters, to review its structure
  • submit_structure_reviewa structure review, re-checked and measured by the server
  • report_used_chunksthe excerpts that actually helped

Connected in a minute

Generate a key from the “API & MCP” screen, choose its spaces and permissions, then paste the configuration into your client. The same index also powers the REST API, for your integrations.

{
  "mcpServers": {
    "ts-knowledge": {
      "type": "http",
      "url": "https://api.ts-services.com/v1/knowledge/mcp",
      "headers": {
        "Authorization": "Bearer YOUR_API_KEY"
      }
    }
  }
}
  • Claude
  • ChatGPT
  • Gemini
  • Mistral
  • IDE agents
  • Your own code
API & MCP screen: the MCP server URL and the configuration to copy into your AI client.

Five steps, and your AI knows your documents

  1. Create a space

    A project, a client file, a document collection: the boundary that will decide who sees what.

  2. Add your documents

    Files, pasted text, a website or a synced source. Indexing happens in the background.

  3. Test it

    Ask your real questions in the Playground: excerpts, scores and sources.

  4. Turn on knowledge

    Optional: entities, facts, relationships, a timeline and the Snowflake, drawn from your documents.

  5. Connect your AI

    One key, one MCP URL, and your agent works with your knowledge.

Spaces screen: each space shows its type, its documents and its collections.
Your spaces, and what each one contains.
Documents screen: the list of documents with their indexing status.
Indexing runs on its own — and you can watch it happen.

Your data stays yours

Strict isolation

Every request is filtered by account and by space on the server — search, profiles and evidence included. An ID supplied by a client never grants access on its own.

Your model keys

AI provider keys are encrypted at rest, tested before they are accepted, and never come back out of the API.

Clear quotas and usage

Storage, documents, searches: your plan's limits are always on display — not discovered the moment you get blocked.

Who is it for?

The same three layers, in very different lines of work. Each use case shows, with concrete examples, what search, structured knowledge and the Snowflake change — and how your agent puts them to work.

Series authorsYour series bible, kept up to date on its ownBooks, drafts and notes from the same universe: who knows what, what contradicts what, what is still missing — without ever spoiling what comes next.See the use case Online stores and small businessesYour whole store, on a single mapProcedures, rules, decisions, systems: a support agent that cites the right procedure — and the terms of sale that contradict customer service, finally in plain sight.See the use case Law firms and legal departmentsEvery matter walled off, the law as of the right dateLetters, expert reports and court decisions split along their natural structure, French law as a shared library, and a timeline for every matter.See the use case Software companies and product teamsYour docs and code, queryable from the IDEA synced GitHub repository, its services and its architecture decisions, within reach of every developer's agent.See the use case

Ready to plug your AI into your own knowledge?

Create a space, add a few documents, ask your first question — then give the same key to your agent.

Create an accountI already have an account