🧭Co-Pilot: from a sentence to a Canvas

How it works#

1. 1

   Describe the process

   On the Dashboard, write what you want to automate in the Co-Pilot box, or upload a document: who starts it, which systems are involved, what it must deliver. Analyze my process opens the Process X-ray.

2. 2

   Review the analysis

   The Solutions Engineer (Claude) answers with the Sub-Agents and what each one owns, their tools and the systems they reach, the Company Memory they read and the rules your text implies, each one highlighted in your own words. Accept or reject pieces, choose how the AI works, answer its questions, ask for changes as often as you like.

3. 3

   Build

   Nothing is drawn before you decide. Build on the Canvas turns what you kept into the workflow, with every step shown: drafting, checking the Canvas rules, laying out, saving to your account.

4. 4

   Audit on the Canvas

   The Canvas opens with every node placed, linked and filled in: behavior rules, tool descriptions, input schemas, credential names, and the sentences each node came from. Review it, run a Simulation, then go Live.

Start from a template: Customize with AI#

Every template in Templates opens in a Spotlight: the workflow drawn read-only as it will look on the Canvas, what it delivers, and which of its tools you already connected. From there, Clone into a new workspace copies it as it is, and Customize with AI hands it to the Co-Pilot first.

1. 1

   Tell it about your business

   What you sell or do, the systems you use and what must change. The Co-Pilot starts from the template's agents, tools and behavior rules, not from a blank page.

2. 2

   Review what changes

   The proposal is the whole adapted workflow, with What changes from the template listed first: a new step, a rewritten rule, a system swapped for another.

3. 3

   Approve

   The Canvas opens with the adapted workflow, and its Briefing records the template it came from.

Glass box: nothing hidden#

• A summary of the Solutions Engineer's reasoning is shown under each answer (How the Solutions Engineer reasoned).
• The proposal is structured: agents, tools, systems, documents and assumptions, not a wall of text.
• Every repair made while building is listed: a link turned the right way round, a node attached to a parent, a function renamed to a valid name.
• The result is an ordinary workflow: every agent, rule and tool is a node you can open, change or delete.

Live and Simulation#

Co-Pilot Memory#

The Co-Pilot can remember facts about your organization from one design to the next, so you don't repeat your policies and systems every time. It remembers only what your team saves, never the conversations.

• What it knows. The facts listed under Co-Pilot Memory on the Dashboard and, without storing anything, the systems you connected and your last five designs approved with the Co-Pilot.
• How facts get in. Type one on the Dashboard, or accept a suggestion: when you mention something lasting about your organization, the Co-Pilot offers Remember for future designs? under its answer. Nothing is saved without a click.
• Your team is in control. Everyone sees the facts; builders and admins add, edit and delete them; admins switch the memory off or forget everything.
• Never secrets. Text that looks like a password or key is refused. Keep credentials in the Vault.
• Small on purpose. Up to 50 facts of 240 characters. It travels as text with the Co-Pilot steps that already run, on your key: no extra request and nothing new to host.

What is sent and what is kept#

• The conversation lives in your browser, kept there for up to 7 days so a reload or a closed tab doesn't lose it (Continue your analysis on Home). Start over, building the workflow or signing out removes it. Each step sends it to our server, which forwards it to Anthropic with your key and keeps nothing of it.
• Only token counts are recorded, for the Usage page.
• The design never contains credentials: Connectors only name the environment variable that will hold each one.
• The workflow is saved to your account like any other project (or kept in this browser when your plan's workspace limit is reached), with its Briefing: your first message and the proposal that was approved, who approved it and when. Open it from the Canvas top bar. The conversation itself isn't part of it.

Descriptions that work#

• Name the systems: “Stripe invoices”, “Intercom conversations”, “our Postgres orders table”.
• Say what starts the process and what it must deliver: “every Monday… a report on Slack”.
• Say where a person must approve: payments, customer replies, production changes.
• Mention the company rules the agents must follow: a policy, a price list, a tone of voice. They become Company Memory.