Knowledge management for consultancies

An opinionated framework for company knowledge.

nbox reads your meetings, threads and files, then files what it learns into one cited, version-controlled tree. You get a page for every client, person and project, and your team and your agents both work from it.

Request access See how it works

Incoming

nbox

acme-advisory/main

·······waiting for sources…
Works with Granola Slack Google Drive Notion Linear Todoist Claude ChatGPT Mistral Kimi

Your firm learns something on every call. Almost none of it gets written down where anyone can find it.

Lost in the call

The sponsor changed and the budget moved. It's in a transcript nobody will reread.

Buried in threads

Scope decisions get made in Slack and are gone a week later. Engagement wikis go stale by the second sprint.

Stuck in one head

When the partner who knows the client is out, so is the context. Your AI tools never had it in the first place.

How it works. Four steps, and nobody has to file anything.

01

Connect

Link Granola, Slack and Drive, or forward an email. Originals are stored untouched.

02

Extract

Each source becomes a note. Facts, commitments and the people involved are pulled out with dates.

03

File

Facts land on the right client, person, project or topic page, with a citation. Every change is a commit.

04

Use

Browse it, ask it, export it, or point Claude, ChatGPT or Cursor at it, with the same permissions you have.

01 / Sync

Connect the places work already happens.

You don't have to change how anyone works. nbox listens to the tools your team already uses and turns each item into a note plus cited facts on the right pages.

  • Granola meetings, Slack threads (or just @nbox one), Google Drive files and forwarded email
  • Originals are kept exactly as they arrived, so you can always trace a fact back to them
  • Notion is next, with more connectors after it
Notion soon
Written to the tree
02 / Task discovery

The promises made on calls become tasks.

"I'll send that over by Friday" usually gets forgotten. nbox catches commitments and follow-ups as sources come in and suggests them as tasks. Each one has an owner, a due date and a link to the moment it was said.

  • Suggestions are kept apart from explicit commitments, so nothing is assigned without review
  • Accept a task and nbox sends it to Linear or Todoist
  • Each task links back to its source, so you can see why it exists
Northwind · quarterly review
Listening for commitments…
03 / File tree

Everything maps to a file tree you own.

There's no proprietary database to escape. Your knowledge is plain Markdown in a git repository, with one folder per kind of thing and one page per thing. People can read it, diff it and export it.

  • One fixed shape: clients/ people/ projects/ topics/ sources/
  • Every fact is dated and cited to its original source
  • Every edit, by nbox or a person, is a commit you can review or revert
  • Clone it, zip it, or open it in any editor
clients/northwind-logistics.md Export .zipgit clone.md
04 / Built for agents

Easy for agents to navigate.

Agents don't need another vector search that guesses. nbox gives them a predictable map: start at the entity page, follow its citations, and read the original only when the exact wording matters.

  • A remote MCP server with OAuth works with Claude, ChatGPT and Cursor
  • Agents get the same per-space access as the person who connected them
  • Answers come back with citations you can click through
connected to nbox · MCP
05 / Your data, your model

Stored in the EU. Inference on the provider you choose.

Originals, notes and workspace data stay in EU data centres. Inference runs on the provider you pick, with your own key. Mark each key as EU or global, then require EU processing when a client asks for it.

  • Anthropic, OpenAI, Google, Mistral or OpenRouter
  • Use separate fast and smart models, and choose which models people can use in Ask
  • No lock-in to one vendor's roadmap or pricing
Settings · AI providers

Storage

Always in the EU
Originals EUEvery source, exactly as received
Notes & pages EUOne git repository per space
Workspace data EUMembers, spaces, connections, jobs
inference

Inference

Your keys, your choice. Click a region to change it.

Opinionated, so your team doesn't have to agree on a system first.

Most knowledge tools give you a blank page and wish you luck. nbox makes the structural decisions for you, and they're the right defaults for a services firm.

Without nboxWith nbox
01Every team invents its own folder structure.One shape: clients, people, projects, topics, sources.
02Notes say what was decided, but not where it came from.Every fact is dated and cited to its original.
03The same mistake gets corrected over and over.Fix a page once and the correction becomes a guideline nbox follows.
04Knowledge is locked inside a vendor's database.Plain Markdown in git. Clone it and leave whenever you like.

FAQ

Common questions about nbox.

Who is nbox for?

Consultancies and other client-services firms, where knowledge is organised around clients, people and engagements and comes from meetings and conversations rather than documents.

What happens when nbox gets something wrong?

Edit the page. You can correct, retract, rename, merge or restore. Every correction is a commit and becomes a guideline and a regression case, so the same mistake is less likely next time.

Who can see what?

Knowledge is organised into spaces with their own members. The web app, the API and connected agents all go through the same access check.

Which model does nbox use?

The one you configure. Bring your own key for Anthropic, OpenAI, Google, Mistral or OpenRouter, and mark each key as EU or global processing.

Can I take my data with me?

Yes. Your notes are a git repository of Markdown files, and the originals are kept as received.

Start with one client.

Connect a single source and watch the tree fill in. It takes about ten minutes.

Request access