Skip to content

Workbench · for VS Code and Cursor

Your AI, working from what your organisation knows.

Workbench is a small extension for VS Code and Cursor. Sign in once, and the AI you already use answers from your organisation’s recorded knowledge, cites it, and tells you when nothing is recorded. New decisions are saved back with your approval, and work is tracked as tickets, without leaving your workspace.

Version 0.0.1 · ii-workbench-0.0.1.vsix · VS Code 1.102 or newer, or Cursor

What changes

Your day, in one place.

Today the work is split. The question goes to an AI that forgets, the answer sits in a filing system, the decision lands in a document, the task lives in a project tool, and the context stays in your head. Workbench brings all of it to where you work, and what you learn stays with the organisation.

What you do in Workbench

  • Ask, and get your organisation’s answer.

    Cited to the document, the author and the date, from the knowledge your organisation has recorded.

  • Decide, and it is recorded once.

    Say yes when your assistant offers, and the decision becomes a fact that everyone’s AI reads from.

  • Hand work to your AI, and see it tracked.

    The work goes onto the Coordinator as a ticket, with who is doing it and what was decided.

  • Pick up where the last session ended.

    Every session starts from what the last one learned, whether it was yours or a colleague’s.

  • Get help in the same conversation.

    Support is a connected tool. Ask, and the question reaches us.

What you stop doing

  • Pasting the same background into your AI every morning because it forgot.
  • Searching four systems for the document that holds the answer.
  • Writing a decision into a document nobody will find again.
  • Updating the project tool by hand after the work is done.
  • Asking the one colleague who happens to know.
  • Re-checking what was already settled last week.

None of it was the work. It was the cost of the work being split across places that do not talk to each other.

Why you are faster.

The time went on the gaps between places: finding the document, re-explaining the background, re-checking what was settled, updating the tool afterwards. In Workbench the answer, the decision and the work are already in the same conversation, and the next session starts with all of it. Your AI stops working from a blank page, so you stop paying for the same context twice.

Why the output can be trusted.

Every answer carries its source, its author and its date, so it can be checked in seconds rather than taken on faith. When nothing is recorded, your assistant says so, which is how you know the rest is real. Decisions are recorded once and read by everyone, so two people no longer work from two versions of the truth. And the work is on the record, so “done” is something you can show.

What it is

What it looks like.

Your assistant gains three things the moment you sign in: your organisation's knowledge in Cornerstone, its work in the Coordinator, and support. Nothing about the assistant changes. What it knows does.

residency-reviewWorkbench · connected
Claude·Cornerstone, Coordinator and Support connected

What is our position on data residency for EU clients?

EU client data stays in EU regions, and Frankfurt is the default. The residency addendum has to be attached to any EU contract. That rule was set by Legal in March, and the default region was an architecture decision in June.

  • Rule · EU client data stays in EU regions · Legal · 12 Mar 2026
  • Decision · Frankfurt is the default EU region · Architecture · 4 Jun 2026
  • Reference · Data residency addendum, §4 · Contract template v3

Is there a rule for the new Milan region?

Nothing is recorded for Milan. The EU rule predates that region, so it may still mean Frankfurt only. Legal would need to confirm rather than assume.

Once Legal has decided, shall I record it in Cornerstone?

Yes, record itNot now

Ticket opened on the Coordinator: Confirm residency rule for the Milan region · Legal

Ask anything about the work, or the organisation…
Illustrative. Your assistant answers with the facts cited, says so when nothing is recorded, and offers to save what you decide.

How it works

Set up in a few minutes.

Four steps to get going, and two more if you want your assistant to work the way the platform is designed for.

  1. 01

    Install the extension.

    From the Marketplace, or with the download below if your machine is locked down. Cursor installs the same file.

  2. 02

    Have your AI assistant ready.

    Claude, which we recommend, or Codex, signed in with your own account. Workbench adds no model of its own.

  3. 03

    Sign in to Instructed Intelligence.

    One sign-in, in your browser. It is the only sign-in you ever do, and it reaches everything.

  4. 04

    Connect your tools.

    One command wires Cornerstone, the Coordinator and support into your assistant. They appear in your next chat, so start a new one.

  5. 05OPTIONAL

    Add the knowledge guide.

    A short note in your project's guide file. It tells your assistant to check recorded knowledge before answering and cite it, to say so when nothing is recorded, to offer to save new decisions, and to treat work as tickets.

  6. 06OPTIONAL

    Simplify the layout.

    One command turns VS Code into a chat-forward workspace. A second puts every setting back exactly as it was.

What it does for you

Everything the platform does, from one place.

This is where the whole platform shows up in your day: answers you can check, decisions that stay made, and work that stays visible.

  • Answers from what your organisation actually knows.

    Ask in your own words. Your assistant reads the recorded facts in Cornerstone and cites them, with the source, the author and the date, so you can check any answer in seconds.

  • It tells you when nothing is recorded.

    When the knowledge is not there, your assistant says so instead of guessing. That is what makes the rest of its answers worth trusting.

  • Decisions are saved back, with your yes.

    When you settle something in a chat, your assistant offers to record it. Say yes and it becomes a fact the whole organisation reads from. Nothing is saved without you.

  • Work becomes tickets, from the chat.

    Ask for a piece of work and it goes onto the Coordinator as a ticket, with who is doing it and what was decided, so the next session starts where this one left off.

  • Help without leaving your work.

    Support is one of the connected tools. Ask your assistant, and the question reaches us.

  • Your AI stays yours.

    You bring Claude or Codex on your own subscription. We add no model and no cost, we never see your credentials, and we never route your prompts.

Workbench is one way to reach the platform. The same knowledge and the same work are there for every model the platform supports. See the whole platform.

Get it

What you need, and the download.

  • VS Code 1.102 or newer on macOS or Windows, or Cursor with Claude Code.
  • A Claude or Codex account.
  • An Instructed Intelligence account. Two people at no cost.

Workbench adds no AI of its own, replaces nothing you already use, and adds no login. You bring your assistant, keep your tools, and sign in once.

Download

Workbench 0.0.1

ii-workbench-0.0.1.vsix

To install the downloaded file, run this in a terminal. Cursor users replace code with cursor.

code --install-extension ii-workbench-0.0.1.vsix

If something goes wrong

Two things to know.

If your AI says it cannot reach your knowledge, start a new chat. The tools join the next conversation, not the one already open.

For anything else, write to enquiries@instructedintelligence.com. A person replies.

Advanced: the commands
  • II Workbench: Sign in

    Signs you in to Instructed Intelligence in your browser.

  • II Workbench: Connect my tools

    Wires Cornerstone, the Coordinator and support into your assistant.

  • II Workbench: Add the knowledge guide

    Writes the behaviour guide into your project's guide file.

  • II Workbench: Simplify the layout

    A chat-forward window. Restore puts everything back.

  • II Workbench: Show connection status

    What is connected, and as whom.

Privacy

In plain English.

  • We never see your Anthropic or OpenAI credentials, and we never route your prompts.
  • Your session lives in your operating system's keychain, never in a file.
  • The session token goes only to Instructed Intelligence services, over HTTPS.
  • The extension holds no secrets of its own. There is nothing in it to steal.
  • Signing out removes the connections.
  • The extension collects no telemetry.

Put your AI to work on what your organisation knows.

Get started with a workspace at no cost, or book a call if you are rolling it out across an organisation.