# ThinkData — A data-intelligence platform turning raw data into knowledge assets

> ThinkData is a data & knowledge platform by ThinkAlike covering eight capabilities: multi-source collection, cloud storage governance, curation & quality, document knowledge base (hybrid retrieval + Q&A), entity recognition & knowledge graph, asset catalog & lineage, open API service, and SN licensing & online update — turning scattered data into reusable enterprise knowledge assets. Backend FastAPI + SQLModel + SQLite, DuckDB for curation, Vue3 + Element Plus front end.

- Product: ThinkData / ThinkData
- Brand: ThinkAlike (Beijing ThinkAlike Technology Co., Ltd.)
- Licensing: SN one-machine-one-code, 90-day free trial and offline on-premises deployment (contact contact@thinkalike.com.cn to book a demo)
- Page: https://thinkalike.com.cn/en/thinkdata.html

## Core capabilities
- **Multi-source collection** — Local-directory, URL-fetch and DB-snapshot collectors with scheduled dispatch and job mutex locks build a reliable data foundation.
- **Cloud storage governance** — Object-storage intranet archiving with sha256 dedup, lifecycle tiering and signed downloads; capacity and cost are visible, staging overflow degrades gracefully.
- **Curation & quality** — In-process DuckDB curation produces Parquet with encoding normalization, row dedup and type inference; a built-in quality rule engine grades results pass/warn/fail.
- **Document knowledge base (hybrid retrieval + Q&A)** — Parses docx/xlsx/pptx/pdf into a knowledge base with FTS5 Chinese full-text and vector hybrid retrieval (RRF fusion) and knowledge-base Q&A.
- **Entity recognition & knowledge graph** — Dual-mode NER (rule dictionary + LLM) extracts entities and relations, exports JSON/CSV/Cypher, and renders a force-directed graph you can reason over.
- **Asset catalog & lineage** — Datasets, knowledge bases and services are cataloged together with auto-registered upstream/downstream lineage and cycle detection, under controlled sensitivity and permissions.
- **Open API service** — API keys (SHA256, scopes, rate limits, revocation) expose search, Q&A and dataset-snapshot endpoints for upstream AI apps and digital employees.
- **SN licensing & online update** — One-machine-one-code SN activation, a 90-day free trial and a read-only demo mode when unlicensed; one-click online update with automatic rollback on failure.

## Highlights
### Store data — and mine it into knowledge
Built-in document parsing, hybrid retrieval, entity/relation extraction and graph pipelines mine unstructured documents and structured tables alike, forming a searchable, answerable, reasoning-capable knowledge asset. When embeddings or the LLM are unreachable it falls back to full-text and marks it honestly, never breaking ingestion.

### Make data a reusable, value-appreciating asset
Processed results are registered as data assets — searchable, grantable, reusable — with auto-registered collect → dataset → KB → service lineage and cycle detection, linked to the suite's knowledge bases and digital employees via the open API so data keeps appreciating in use.

## Interface preview
Screenshots taken from ThinkData's production UI with real data (relative path `images/thinkdata/`):
- Knowledge base — `images/thinkdata/01-kb.png`
- Hybrid search — `images/thinkdata/02-kb-search.png`
- Knowledge graph — `images/thinkdata/03-graph.png`
- Asset catalog — `images/thinkdata/04-assets.png`
- Open-service keys — `images/thinkdata/05-keys.png`

## Use cases
Enterprise data governance · Industry knowledge graph · Document assetization · Q&A data supply · Private knowledge base

## FAQ
**What is ThinkData?**

ThinkData is a <b>data & knowledge platform</b> by ThinkAlike spanning multi-source collection, cloud storage governance, curation & quality, document knowledge base, entity recognition & knowledge graph, asset catalog & lineage, open API service and SN licensing & online update.

**Is there a free trial? How is it licensed?**

Yes. ThinkData uses <b>SN licensing</b>: after installation you can claim a <b>90-day free trial</b> online and activate with a <b>one-machine-one-code SN</b> before it expires; unlicensed it runs in a <b>read-only demo mode</b> (readable, not writable) — no data loss.

**How does it upgrade safely?**

One-click online update from the SN platform: the install script <b>backs up code and database</b>, runs a health check after replacement, and <b>automatically rolls back</b> to the previous version on failure.

**What stack does it use?**

Backend <b>FastAPI + SQLModel</b> with <b>SQLite</b> for metadata/runtime; <b>DuckDB</b> for curation producing Parquet; retrieval is <b>FTS5 Chinese full-text + vector (numpy, optional pgvector backend) fused via RRF</b>; front end <b>Vue3 + Element Plus</b>.

**How does it relate to the knowledge base?**

ThinkData processes data into reusable knowledge assets that, through the open API, link with <b>ThinKM</b> and the digital employees for retrieval.

**Does it support private deployment?**

Yes. ThinkData ships as an <b>offline, on-premises</b> intranet deployment; data never leaves your domain and <b>a single 4GB server is enough</b>. The hosted demo is coming online — for a demo or deployment plan contact <b>contact@thinkalike.com.cn</b>.