User guide

The Workspace lists the projects you can access in the selected organisation. You only see projects you have been added to, unless you are an organisation administrator or executive, who see all projects.

Each project has milestones, risks, issues, decisions, dependencies, work items and reports. Every list can be searched, filtered, sorted and paged. Open a record to see its owner, status, dates, evidence, related records and audit information.

Whether you can create, edit or delete depends on your role on the project. If a button is missing, you do not have permission.

The Chat tab lets you ask questions about the project. The steward only sees information you are allowed to see, and it remembers the earlier messages of the same conversation (never another conversation or project). It can propose changes, but it cannot make them: each proposal goes to the Approvals tab as its own item for a person to decide. Conversations are saved; you can rename, archive and delete them.

Answers keep verified facts, assumptions, recommendations and the evidence still required clearly apart. Project registers are shown to the steward as totals plus a prioritised list (overdue, blocked, critical, upcoming); when a register is too long to list in full, it is told so and never treats the first rows as the whole picture.

Chat requires an administrator to configure the OpenAI integration. Replies are not streamed; they appear when complete.

To plan a project that doesn't exist yet, use New project setup (organisation admins and project managers). It starts a separate conversation with no project attached, so none of your existing projects' information is used. The steward drafts a charter, governance, a phased plan and a RAID log (risks, assumptions, issues, dependencies), labelling every assumption and listing the evidence still needed. Nothing becomes a project until you click Create this project, which creates it, starts a fresh conversation for it and carries over only your draft.

The steward also returns a structured draft: charter and objectives, scope and exclusions, lifecycle phases, milestones, work breakdown, dependencies, risks and assumptions, decisions and approval gates, stakeholders, required specialists, procurement packages, cost categories, evidence requirements and immediate next actions. You review it, tick the items you want, and only those become records (milestones, work items, dependencies, risks, proposed decisions). Items you don't tick are never created, creating twice does nothing, and every created record is marked as coming from a draft with no evidence yet. Dates are not filled in and risk scores are the steward's estimate.

If you ask for something new while another project is open, or name a different project, the steward asks which you mean instead of guessing: start a new project, open the other project, or apply the request to the project that is open.

When a question depends on current external facts (laws, permits, regulations, standards, professional roles, costs, local processes), the steward searches the web and shows its sources with the date they were retrieved. It prefers official and primary sources and separates general guidance from requirements verified for your jurisdiction. It never claims the list is complete, and it is not legal, planning, building-control or regulatory advice.

If the search is unavailable or returns nothing citable, the reply says plainly that current requirements are UNVERIFIED and tells you which authority to check.

Project managers can queue a background supervision run with the Workspace Agent. The app shows its status (queued, in progress, suspended, completed, failed) and a link to open the ChatGPT conversation. The agent's reply is not available inside Project Steward.