Start a Project
Describe what you want to accomplish; review a proposed project, with the reason for each part; create it only when you confirm.
Updated October 8, 2026 · 3 min read
Where to start
Start a Project is the Dashboard's main button. On Projects, New project → From an objective opens the same page; Empty project is the plain form with a name and a description. Before you have any workflow or connection, the Dashboard's text box starts a project too.
Write what you're trying to accomplish, in your own words, or upload a .txt, .md or .docx. For example: “I want to automatically monitor our website, identify problems and notify our development team.”
The proposal
The Co-Pilot answers with a proposed project, as cards:
| Card | What it shows |
|---|---|
| Project | A name, the purpose, the objective, and (when your text says so) the expected outcome and how it's measured. |
| Recommended agents | Who does what. Each agent's work method stays Not chosen yet until you choose it. |
| Workflow steps | The agents in order, with the steps where a person approves. |
| Required capabilities | What the project needs the connected services to do (for example, *Monitor errors* in Sentry), whether it reads or changes data, and where it stands now: On, Off, Not connected or Runs only in Claude Code. |
| Required connections | The services it uses and their state. Local tools such as Playwright run in the Claude Code export, not in platform runs. |
| References | The documents the agents read. They're created empty: you add their content later. |
| AI provider | Live runs use your organization's Claude connection; Simulation always works without it. |
| Permissions | What still has to be connected or switched on, and whether your role can do it. What you can't set up says Needs an admin. |
| What to test | A Simulation first, then a Live test. |
Why each part is there
Every item says where it came from: your words (quoted exactly, never paraphrased), Drafted by AI when the Co-Pilot suggested it on its own, or Rule in Simulation, where the proposal is the closest ready-made template and the goal map, not a design made for you.
Edit Proposal
Edit Proposal opens the Process X-ray review with the project's fields at the top: rename the project, rewrite its purpose and objective, keep or reject each agent, tool, rule and document, choose each agent's work method and answer the Co-Pilot's questions. On the proposal itself you can also leave a capability out. Nothing is created while you edit.
Create Project
Create Project first lists exactly what it will create, and does it only after you confirm:
- the project, with its purpose and objective;
- one Draft workflow from what you kept;
- the capabilities it needs that are already on, in the project's list.
Then the project's page opens, with its health.
What is sent and what is kept
In a Live proposal, Claude receives your text, the names of the capabilities and goals, which services your organization has connected and switched on (names and states, never a credential) and, when the setting allows, the Co-Pilot's memory facts. Never another project, a Reference's content or a run's content. The conversation stays in your browser.
The project keeps its name, purpose and the three objective fields, which you can change in the project's Settings.