🔬Process X-ray: your text, analysed

How it works#

1. 1

   Describe or upload

   On the Dashboard, write the process in the Co-Pilot box or Upload a document (.txt, .md or .docx, up to 5 MB, read in your browser). Analyze my process opens the Process X-ray with your text.

2. 2

   Watch the preview

   While you type, a Preview — no AI marks decisions, checks, people, systems and actions with word lists in English and Portuguese. It costs nothing and never reaches the AI.

3. 3

   Analyze

   Analyze sends your text to the Co-Pilot: one step on your organization's Claude key. It answers with the pieces of the workflow and, for each one, the words of yours that asked for it.

4. 4

   Decide, then build

   Accept or reject each piece, choose how the AI works for each Sub-Agent (work methods), answer the open questions, then Build on the Canvas.

Reading the highlights#

• Colour, icon and number. Each kind of piece has a colour (Sub-Agents, actions, systems, documents, a person approving, behavior rules) and every highlight ends with the number of its item in the list. Colour is never the only sign.
• Hover or focus an item and its words are outlined in the text. Click a highlight or a number to go to its item.
• Overlaps. Words that belong to more than one item show a count (×2) that lists every item.
• Not used. Sentences nothing came from get a dotted underline, with a counter. Read them: something you need may be missing from the design.
• Suggested by the Co-Pilot (not in your text). Items the Co-Pilot added on its own carry this label instead of a quote.
• Only your words. Every quote is checked on our server against what you wrote. A quote that doesn't match (a paraphrase) is left out, and the page says how many were.

Accept, reject, answer#

• Everything is accepted until you say otherwise. Reject leaves an item out of what gets built. A rejected Sub-Agent also takes its actions, its behavior rules and the approval before it, and the page lists them before you build.
• Decisions, checks, ways out of a failure and records become behavior rules of the Sub-Agent (or the Root Orchestrator) that follows them. A step where a person approves becomes a Human Approval node.
• Missing information comes as questions with an answer field. Update analysis sends your answers; they appear under your text as *Your answers* and can be highlighted too.
• Ask for a change works like the Co-Pilot chat: say what to add, drop or do differently.

The process checklist#

Six checks computed from what you accepted, not from an AI's opinion:

• Decisions have clear rules.
• Results are checked.
• Failures have a way out.
• A person approves where money or customers are affected.
• Decisions are recorded.
• Every Sub-Agent has a work method you chose.

On the Canvas#

• Each node keeps the sentences it came from: Why does this exist? in the inspector, or *Suggested by the Co-Pilot* when nothing in your text asked for it.
• The Briefing on the Canvas records your first message, what you approved, what you left out on purpose and the work methods you chose.

Without a Claude key#

In Simulation the preview keeps working, but nothing analyses your text: Analyze offers the closest ready-made template, labelled *Simulation · no AI*, with its items and no quotes. You can still reject items and choose work methods before building. Real highlights need a Claude key (Connections).

What is sent and what is kept#

• An uploaded document is read in your browser: only its text goes into the box and from there, inside the analysis step, to Anthropic with your key. The file never reaches our server.
• The analysis travels like any Co-Pilot step: through our server to Anthropic, and isn't kept. Your text and your choices stay in this browser (up to 7 days) so a reload doesn't lose them.
• Each node keeps up to 6 source sentences of 300 characters with the workspace, until the workspace is deleted. They're never sent in runs and never exported.