Sign in

Taraz Technologies Knowledge Base

Turn documents into a knowledge graph

Drop a document or a zip bundle, pick its domain and source quality, and watch every pipeline stage live — from semantic chunking to the vector store.

entities
relationships
documents
images

Document

Drag & drop an .html / .htm / .docx / .pdf file — or a .zip bundle (documents + their image folders, e.g. enhanced_diagrams/) — here, or browse.

Metadata

7 / 10

Off = images indexed with their alt text (cheaper, current default).

Pipeline

live · refreshes automatically
No jobs yet — ingest a document to see it move through the pipeline here.

Knowledge base

DocumentTypeDomainAccessReviewQualityChunksEntitiesRelationsImages
The graph is empty — ingest your first document.

Chats

Conversations synced from a connected account. A chat is not a file — it has a roster, a date range and a speaker on every line — so it gets its own view. One chat is one document, however many of its members connect.

Mail

Outlook threads synced from a connected account. Mail is opt-in by correspondent: nothing is read or ingested unless you tick the person, and every thread stays readable by its mailbox owners alone. One thread is one document, however many of its participants connect.

SharePoint

Files synced from the document libraries you select. Ingestion is opt-in per library: pick a library, choose who may read its documents, and untick any file that should stay out. The same file at several paths becomes one document — copies are detected and parked, never re-processed.

Your Claude, connected

A personal connector lets your own Claude search and ask this knowledge base as you — it sees exactly the documents your account can see, nothing more. Create one, add it to Claude, done.

New connector

The credential is shown once, right here — it is never stored on the server and cannot be shown again. Treat it like a password.

Your connectors

LabelCredentialCreated ExpiresLast used
none yet
kills every connector and signs you out on every device — there is no narrower revoke for stateless credentials

App connectors

Connect your accounts — Gmail, Drive, Teams, Microsoft 365, ClickUp, Odoo — the same way connectors work in Claude. The ingest agent (coming next) will only ever act through the permissions you set here: read is on by default, write is off until you turn it on, and disconnect kills the credential.

loading…

Graph editor

Correct or remove what extraction produced. Edits write straight to the knowledge graph and re-sync Neo4j; retrieval picks them up automatically. Renaming an entity keeps its old name as an alias so existing queries still match. Adding brand-new nodes or edges isn't offered here — answers are only trusted when grounded in a document quote.

EntityTypeDescriptionRelsDocs

Usage & cost

last 30 days
$0.00 no usage recorded yet

Daily spend

Estimated from token usage × a per-model price table — not a billing figure. Neither OpenAI nor Gemini exposes a remaining-credit balance through an API key, so this shows our metered spend for ingestion (extraction, resolution, embeddings). Calibrate the price table to your real rates via the PRICE_TABLE_JSON environment variable.

Org

Who can read what

A person reads their own node and everything above it — that is membership, and it is automatic. Reading below a node is a manager privilege and must be granted deliberately. Projects sit outside the tree: membership is by grant only, so they cross departments. A document can carry several tags at once.

organisation department team person project home (the tree) also a member
Click a node for details · click a person, then a department, to put them in it

Policy

Org-wide rules that govern what the knowledge base may do — who can read what, and what mass ingestion is allowed to produce. Admin-only; every change is recorded in the ledger with who made it.

Access

SharePoint mass ingestion

What a SharePoint ingestion is allowed to produce. Full builds the knowledge graph from every file (entities & relations — thorough, slower, costs model time per document). Vector only makes files searchable and citable without graph extraction — fast and near-free, ingested automatically. User's choice lets whoever runs the ingestion pick per selection.

API keys

Create a key

A new key may query and nothing else. Tick only what this integration genuinely needs — a key that shares no resources still gets grounded, verified answers, just without the material behind them.

This is the key's identity for access control (principal key:<name>), so it must be unique.
The sync bot reads every connected user's staged mail and chat and files it as its owner. Only the auto-ingestion engine should ever hold it.
Purge is never available to a key — it destroys provenance irreversibly and stays an admin action.
All unticked = answers only, no sources. The answer is still retrieved and still verified; nothing is invented.
loading…
Scope comes from the org tags (org / dept / project) — domains are filters, not access control. Tick none to grant nothing yet; you can change scope any time by clicking the key.

Existing keys

Click a key to configure what it can do and what it may see.

Users

Add a user

Admins can add domains, delete documents and manage users. A normal user may ingest into existing domains and view the graph.

UserRoleCreated