01 · Vector
Semantic
Embeddings find passages that mean the same thing, even when they share no words.
360° GraphRAG · Knowledge Studio
Turn documents into knowledge that connects.
MENTOR extracts, chunks, embeds and connects everything your team writes into a living knowledge graph — then answers from all 360° of it, with every claim cited.
- Self-hosted on your own server
- Every answer cited
- Organisation-scoped libraries
Reads the formats you already have · runs on the stack you trust
- Word
- PowerPoint
- Excel
- HTML
- Markdown
- CSV
- Web URLs
- Self-hosted
- Docker Compose
- PostgreSQL + pgvector
- S3-compatible storage
- Built-in work queue
- OpenAI
- Anthropic
- Ollama
01What is 360° GraphRAG
Search reads text. MENTOR reads the connections.
Classic RAG retrieves a few similar paragraphs. 360° GraphRAG looks at your library four ways at once — meaning, wording, the entities involved and the themes around them — and fuses them into one grounded, cited answer.
02 · Keyword
Lexical
Exact terms, codes and names are matched too, then fused with the semantic hits.
03 · Local graph
Entity neighbourhoods
Entities in the question pull in their relationships, source chunks and communities.
04 · Global
Community summaries
Clusters of related entities are summarised, so broad questions get whole-library answers.
SemanticLexicalLocalGlobalone cited answer
Fused and re-ranked, then written up with numbered citations to the exact passages.
- retrieval modes
- 6
- Vector, hybrid, graph, local and global — or all fused as 360°.
- context
- 360°
- Meaning, wording, entities and themes behind every answer.
- cited answers
- 100%
- Every claim links to the exact passage and page it came from.
- to a first graph
- <5min
- Drop in a folder of documents and watch the graph build itself.
02How it works
From upload to answer in seven steps.
Every document runs the same durable pipeline on your own server — and you can watch each stage happen.
Step 01
Upload
Drop in PDF, Word, PowerPoint, Excel, HTML, Markdown or CSV — or paste a URL. Originals are kept in your own object storage.
Step 02
Queue
Every document joins a work queue in your database. Workers pick it up, track each stage, retry on failure and resume after a restart.
Step 03
Extract & chunk
Text and structure are pulled out, then split into token-sized chunks that remember their page numbers.
Step 04
Embed
Each chunk becomes a vector in PostgreSQL + pgvector, ready for semantic and hybrid search.
Step 05
Graph
A model reads every chunk for entities and relationships, and duplicates merge into one library-wide graph.
Step 06
Communities
Related entities are clustered at two levels, and each community gets a written summary and key findings.
Step 07
Ask
Questions run across every retrieval mode and come back as one answer, with numbered citations to the source.
03Sources
Every format. One graph.
Board packs, handbooks, spreadsheets and web pages all flow into the same core — extracted, connected and ready to question.
Mentor Graph
- PDF
- DOCX
- PPTX
- XLSX
- HTML
- MD
- CSV
- URL
Page-aware extraction
Chunks keep their page numbers, so every citation opens at the right page.
Links, fetched for you
Paste a URL and MENTOR fetches it server-side, then queues it like any upload.
Re-process any time
Change the schema or the models, then re-process a document or rebuild communities with one click.
04360° retrieval
Six lenses. One answer.
Pick a lens to see how MENTOR looks at your library — or let 360° run them all and fuse what they find.
Question
How exposed is Nimbus GA to supply risk?
360° lens
Hybrid + graph + local + global, fused.
Best for: Anything — it's the default.
Hover or tap a lens · every few seconds they converge into one answer
05Live demo
Explore a real knowledge graph.
A sample library — a fictional company's strategy papers — as MENTOR extracts it. Hover to trace connections, click an entity to inspect it, switch between types and communities.
Loading graph
- Click an entity for details
- Drag to pan
- ⌘/Ctrl + scroll or pinch to zoom
06Features
Everything a knowledge team needs. Nothing it doesn't.
Libraries & organisations
A library for every team
Each library has its own documents, schema, graph and settings, scoped to a Clerk organisation. Admins shape it; members explore and ask.
Strategy 2026142 documents · acme-orgIndexedEngineering handbook318 documents · acme-orgProcessing 62%Customer contracts57 documents · acme-orgIndexed- Org-scoped
- Roles
- Per-library schema
Graph explorer
Bloom-style exploration
Expand neighbours, find paths between two entities and filter by type or community on a fast WebGL canvas.
- Neighbours
- Paths
- Communities
Graph health
Dedupe without the drudge
Near-duplicate entities, orphans and weak links surface on their own. Merge or dismiss each suggestion in one click.
Nimbus EdgeNimbus Edge gatewayNimbus Edge96%- Duplicates
- Orphans
- Merge
Schema & ontology
Your language, your types
Start from a preset or let MENTOR suggest entity and relationship types from your own documents, then refine them.
- Presets
- Suggestions
- Schema graph
Retrieval details
See why it answered
Every run keeps its passages, scores, modes, tokens and latency — with an optional quality evaluation on demand.
- Run history
- Scores
- Evaluation
Admin & queues
Operations you can actually see
Watch queues and jobs live, retry failures, replay poison messages, tune models and limits — with a full audit trail.
ingest12 runninggraph-extract4 runningcommunitiesidleingest-poison0 messages · replay- Jobs
- Retries
- Audit log
07Use cases
Built for the documents you actually have.
Strategy & board papers
Institutional memory for the boardroom
Years of board packs, strategy decks and risk registers become one graph of decisions, owners and dates.
- Trace any decision back to the paper and page
- Ask what changed between quarters
- Brief new directors with community summaries
Strategy 2026 · 142 documents
Sources
- Board pack Q3.pdf1
- Strategy 2026.pptx2
- Risk register.xlsx3
Ask · 360°
What did the board decide about Nimbus in Q3?
The Q3 Board Review approved the Nimbus investment case 1 and made supplier diversification a Q4 priority 2.
VectorGraphLocalGlobal
08Security & platform
Self-hosted. Kept on your server.
MENTOR runs as one Docker Compose stack on a machine you control: web app, API, workers, database, object storage and an HTTPS proxy. Your documents are never stored anywhere else.
Your server, your data
Original files live in object storage; chunks, vectors, entities and community reports live in PostgreSQL — both in containers on a machine you run.
Organisations built in
Clerk handles sign-in and organisations. Every library, query and job is checked against the caller's organisation and role.
Durable by design
A work queue in the database tracks every stage, retries failures, survives restarts and lets admins replay anything that got stuck.
Accountable
Admin changes land in an audit log, and every answer keeps the retrieval details behind it.
Your server
MENTOR StudioNext.js web app
ClerkSign-in · organisations
API & workers.NET containers
Object storageoriginal files · S3
Work queueingestion pipeline
PostgreSQLpgvector · graph
Your modelsOpenAI · Anthropic · Ollama
- MENTOR StudioNext.js web app
- ClerkSign-in · organisations
- API & workers.NET containersyours
- Object storageoriginal files · S3yours
- Work queueingestion pipelineyours
- PostgreSQLpgvector · graphyours
- Your modelsOpenAI · Anthropic · Ollama
09FAQ
Questions, answered.
The short version of how MENTOR works, where your data goes and why every answer comes with its sources.
Retrieval-augmented generation that looks at your library from every side. Alongside semantic (vector) and keyword search, MENTOR builds a knowledge graph of the entities and relationships in your documents, clusters them into communities with written summaries, and fuses all of it — vector, hybrid, graph, local and global — into one grounded answer.
Mentor is distributed intuition
Your documents already know. Let them answer.
Create a library, drop in a folder and ask your first question — with citations — in minutes.