Company Brain

Give your agents
a company brain.

One workspace that holds what your company knows, and answers questions about it with the file behind every claim. Any agent can ask. Setup takes about ten minutes.

claude code

> What did we promise Northwind about the migration timeline?

Ask the Brain · workspace: Company Brain

Two things were committed. On the kickoff call you agreed to a staged cutover finishing March 14, and the signed plan adds a two-week parallel run before their old system is retired. The contract itself sets no date.

Northwind-Kickoff-Call.txt Migration-Plan-v3.pdf Northwind-MSA.pdf

How a company brain is put together

Three parts. Files go in, Fastio indexes them, and anything that speaks MCP can ask.

What goes in

Contracts and PDFs Call transcripts Specs and decisions Exports and reports Whatever your agents write back
Company Brain

One Fastio workspace, with intelligence turned on

Files index automatically as they arrive
Keyword and meaning-based search across everything
Answers cite the document behind every claim

Who asks

Claude Code Codex Cursor Gemini OpenClaw Your team, in the app
The pattern

People have been building these by hand

Give an agent a file it can read and you have given it memory. That idea has been circulating among the people who work with agents every day for a while now.

BB
Birgitta Böckeler
martinfowler.com

A coding agent has none of this: no social accountability, no aesthetic disgust at a 300-line function, no intuition that “we don't do it that way here,” and no organisational memory.

Distinguished Engineer, Thoughtworks · Apr 2026
AK
Andrej Karpathy
@karpathy

…context engineering is the delicate art and science of filling the context window with just the right information for the next step.

Founding member of OpenAI, ex-Tesla AI · Jun 2025
SW
Simon Willison
simonwillison.net

The key idea here is that coding agents mean we only ever need to figure out a useful trick once. If that trick is then documented somewhere with a working code example our agents can consult that example and use it to solve any similar shaped project in the future.

Co-creator of Django, creator of Datasette · Feb 2026
MH
Mitchell Hashimoto
mitchellh.com

For simple things, like the agent repeatedly running the wrong commands or finding the wrong APIs, update the AGENTS.md (or equivalent). … Each line in that file is based on a bad agent behavior, and it almost completely resolved them all.

Co-founder of HashiCorp, creator of Ghostty · Feb 2026
TL
Tobi Lütke
@tobi

I really like the term “context engineering” over prompt engineering. It describes the core skill better: the art of providing all the context for the task to be plausibly solvable by the LLM.

Co-founder and CEO of Shopify · Jun 2025
YJ
Yichao “Peak” Ji
manus.im

…using the file system not just as storage, but as structured, externalized memory.

Co-founder and Chief Scientist, Manus · Jul 2025

Every quote links to the original. None of these posts are about Fastio, and none of these people are endorsing it. They are describing the practice.

About ten minutes

Have your agent set it up

Paste this into Claude Code, Codex, Cursor, or whatever you use. It will do the whole thing and stop when it needs you.

Setup prompt
Set up a Company Brain for me on Fastio.

1. Connect to the Fastio MCP server at https://mcp.fast.io/mcp/oauth and walk me through signing in.
2. Create a workspace called Company Brain, and turn on intelligence so everything in it gets indexed as it arrives.
3. Upload the files I point you at. If I have folders in Google Drive, Dropbox, OneDrive, or Box, offer to import those too.
4. Create a skill called Ask the Brain. It should always query the Company Brain workspace before drafting, editing, planning, triaging, or changing anything, ask follow-up questions until it has the whole picture, and cite the file behind every factual claim.
5. Add one line to your memory: whenever a task needs company context, use the Ask the Brain skill before doing the work.

Then show me what is in the workspace and ask it one test question. Stop and check with me before anything that costs money or shares a file.

First time connecting? Step by step guides for each agent:

Or do it by hand, in three steps

01

Make the workspace

Create a workspace, call it Company Brain, and turn intelligence on. That is the switch that starts indexing.

02

Put something in it

Start with the ten documents you already rely on. Drag them in, or import the folders you keep in Google Drive, Dropbox, OneDrive, or Box.

03

Point your agent at it

Add the Fastio MCP server, then write the Ask the Brain skill so the agent checks the workspace before it starts guessing.

The one command you need for Claude Code:

claude mcp add --transport http fastio https://mcp.fast.io/mcp/oauth

Questions people ask first

What is a company brain? +

One place your agents can ask about your company. In Fastio it is a workspace with intelligence turned on: files land, index automatically, and any agent connected over MCP can ask a plain English question and get an answer back with citations to the documents behind it.

How long does setup take? +

About ten minutes for the wiring: create the workspace, turn on intelligence, connect your agent to the Fastio MCP server, and write the Ask the Brain skill. You can hand the whole thing to your agent with a single prompt and approve the steps as it goes.

How much has to be in it before it is useful? +

Less than you would think. Ten documents that hold real decisions beat a full archive nobody curated. Add the folder behind whatever you are working on first, then grow it as each answer shows you what was missing.

Does this only work with Claude Code? +

No. Claude Code, Codex, Cursor, Gemini, OpenClaw, or anything else that speaks MCP. There is also a CLI and a REST API for custom setups. The brain lives in the workspace rather than in the agent, so changing tools later does not mean rebuilding it.

Can the agent see things it should not? +

An agent acts as the identity it authenticated with, and answers are filtered by that identity's permissions, so an answer never contains a file the asker could not already open. API keys can be scoped to a single workspace, and every action lands in an append-only activity log you can search.

What does it cost to try? +

Creating an account is free. Doing real work needs an organization, and every organization runs on a plan with a 14-day trial that requires a card. Intelligence, the feature that makes a workspace answerable, is available on every plan.

Give your agents somewhere to ask

One workspace, one connection, and every question your agent has about your company has an answer with the file behind it.