Turn five years of notes into one searchable graph.

Point Nexus at five years of notes, docs and transcripts. It extracts the people, projects and decisions into a temporal knowledge graph you can query in plain English. Every answer is cited back to the note it came from.

Run Nexus
AuthorMariete
AgentNexus
Runtime~ 40 minutes for the first index
IndustryOperations
  1. 1

    Describe the task

    Nexus can take five years of scattered notes, meeting minutes, wikis and drive folders and turn them into one searchable, temporal knowledge graph, no manual tagging, no re-filing. It extracts the people, projects, decisions and dates hiding inside your documents, links them together, and lets anyone on the team ask a plain-English question and get an answer that cites the exact source it came from.

    The more precisely you scope the run, the cleaner the graph. Tell Nexus which drive to index, which entities matter most, and how you want answers cited, and it keeps that structure consistent across every document it reads.

    Index the entire team drive at drive://ops-shared and build one searchable knowledge graph. For every document: - Extract the key entities: people, projects, decisions, vendors, and dates - Link decisions to the meetings and documents they came from - Preserve a temporal ordering so I can trace how a decision evolved When I ask a question, answer in plain English and cite the source file and date for every claim. End with a summary: total documents indexed, entities extracted, and any files you could not parse.

  2. 2

    Give Nexus context

    In a Nexus workspace, point the run at the drive, wiki, or folder tree that holds your history and tell it which entity types matter. Nexus reads everything together, so the same project referenced in a 2021 doc and a 2025 Slack export resolves to a single node instead of two.

    Give Nexus a short glossary of internal names if your team uses codenames or acronyms, "Project Halo" resolving to the billing rewrite keeps the graph from splitting one initiative across three labels.

    Required context
    A source location, a drive, wiki export, or folder tree containing the notes and documents to index
    A list of entity types that matter to you, people, projects, organisations, decisions, dates
    Optional context
    A glossary of internal codenames and acronyms, so aliases resolve to a single entity
    A Connected-mode team vault, so everyone queries one shared graph instead of indexing their own copy
    A priority list of entity types, so the ones that matter most surface first when you query
    ops-shared driveinternal-glossary.mdNotion (optional)
  3. 3

    What Nexus creates

    Nexus works through every document, resolving duplicate references and stitching them into a temporal graph. You get a searchable knowledge base where each answer cites its source file and date, plus a short list of documents it could not parse.

    From Nexus: Indexed 4,120 documents from drive://ops-shared into one temporal knowledge graph.

    4,120Documents indexed
    8,940Entities extracted
    31Files it could not parse
    Top entities by connection count8,940 entities
    EntityTypeLinksFirst seen
    Project HaloProject214Mar 2021
    Renata VossPerson176Jan 2021
    Cardinal LogisticsOrganisation98Aug 2022
    Sample answer with citations3 sources
    ClaimSource fileDate
    Halo shipped after the vendor swapq2-retro-2023.docxJul 2023
    Cardinal replaced the prior 3PLops-review-aug.mdAug 2022

    "31 files could not be parsed, mostly scanned PDFs with no text layer. Want me to run OCR on those and fold them into the graph, or leave them flagged for manual review?"

  4. 4

    Follow-up prompts

    Trace what already depends on a decision

    Once the graph exists, Nexus can trace the dependencies already recorded in it, walk from any decision to everything the documents link to it, so you see what currently depends on it, each dependency cited to the note it came from. It reports the relationships in the graph, it does not forecast what might break.

    Using the graph, trace every project, vendor, and obligation already linked to Project Halo. List what currently depends on it, per the documents, and cite the source note for each dependency.

    Re-index when new notes land

    When new notes or pages land, re-point Nexus at the source and it folds them into the existing graph on your next run, resolving them against the entities already there. There is no proactive sync, you re-run the index whenever you want the graph current.

    Re-index drive://ops-shared and fold any new or edited documents into the existing graph. Show me which decision or vendor entities are new since the last run, each cited to its source file and date.

    Let the team query the shared graph

    In Connected mode the graph lives in a hosted team vault, so anyone on the team can ask it a question and get a cited answer back, no need to re-index their own copy to see what was decided.

    In the Connected team vault, answer: what did we decide about the billing rewrite, and who signed off? Cite the source file and date for every claim.

  5. 5

    Tips and troubleshooting

    Scope the drive before the first run

    Pointing Nexus at a 40,000-file drive with a decade of unrelated attachments produces a noisier graph than a scoped ops-shared folder. Narrow the source first, then widen it once the entity types are dialled in.

    A glossary prevents split entities

    If your team calls one project three different names, Nexus will create three nodes unless you give it a glossary. A short alias list is the single biggest lever on how clean the graph reads.

    Answers cite sources by default

    Every answer names the file and date it came from. If a claim has no citation, treat it as low-confidence, it usually means the fact spanned documents Nexus could not confidently link.

  6. Ready to try it yourself?

    Point Nexus at your drive, name the entities that matter, and get a searchable graph where every answer cites exactly where it came from.

    Run Nexus