ProductFeaturesWork

Projects

Organize your work into projects with documents, libraries, and team collaboration.

Projects are the primary way to organize your work in Fabric AI. They bring together documents, libraries, agents, and team collaboration in one place.

What is a Project?

A project is a container for related work:

  • Documents — PRDs, specs, architecture docs generated by agents
  • Libraries — Uploaded files that provide context for AI agents
  • Artifacts — Pages Fabric Agent writes for this project
  • Conversations — Chat history with agents about this project
  • Settings — Project-specific configuration and templates

Use the project's Coding Instructions tab to share coding rules and connect Claude Code, Codex, or another coding tool. The guide covers repository setup, folder uploads, and reviewing changes.

Creating a Project

Click Projects in the left sidebar.

Create New Project

Click Create Project and provide:

  • Name — Descriptive project name
  • Description — Brief summary of the project
  • Template (optional) — Start from a template

Configure Settings

Set project preferences:

  • Default Document Type — PRD, Spec, etc.
  • Default Agent — Which agent to use
  • Visibility — Personal or shared with organization

Start Working

Your project is ready! You can:

  • Upload documents to the library
  • Start a conversation with an agent
  • Generate your first document

Project Structure

Loading diagram...

Working with Documents

Generating Documents

Open Project

Click on your project to open it.

Start Agent Chat

Click New Document or Chat to start a conversation with an agent.

Describe What You Need

Tell the agent what document you want:

Create a PRD for implementing OAuth 2.0 authentication.
Include user flows, API requirements, and security considerations.

Review and Edit

The agent generates a draft. You can:

  • Edit directly in the rich text editor
  • Request changes by chatting with the agent
  • Accept/Reject suggested modifications

Save Document

Click Save to add the document to your project.

Document Types

Fabric supports various document types:

TypeDescriptionTemplate Features
PRDProduct Requirements DocumentGoals, features, success metrics
Technical SpecTechnical implementation detailsArchitecture, APIs, data models
ArchitectureSystem architecture documentationDiagrams, components, flows
API SpecAPI documentationEndpoints, schemas, examples
ProposalProject proposalsProblem, solution, timeline

Features are no longer created as a document. Recommend, review, and manage Features in the project's Roadmap instead; Features documents created earlier stay readable.

Document Editor

The built-in editor provides:

  • Rich formatting — Headings, lists, code blocks
  • Markdown support — Write in Markdown, render beautifully
  • AI assistance — Inline AI suggestions and rewrites
  • Diff highlighting — See changes clearly
  • Version history — Track all changes
  • Export — PDF, DOCX, Markdown

Project Libraries

Each project has a dedicated library for RAG context.

Uploading Documents

Drag and drop files or click to upload:

  • Supported formats — PDF, DOCX, MD, TXT, code files
  • Size limits — Up to 50MB per file
  • Bulk upload — Multiple files at once

Using Context

When you chat with an agent in a project, it automatically has access to:

  1. All documents in the project library
  2. Previously generated documents in the project
  3. Any explicitly attached libraries

This means the agent understands your project context and can:

  • Reference your existing documentation
  • Maintain consistency across documents
  • Build on previous work

Managing the Library

  • Add files: Upload new documents anytime
  • Remove files: Delete outdated documents
  • Remove duplicates: An item whose content matches another item's is marked Duplicate of …; Remove duplicates deletes the extra copies and keeps the original
  • Living Memory: Files a connected agent pushes from a working tree appear under Living Memory, grouped by folder, with each file's full path under its name
  • Remove a Living Memory folder: Remove folder beside a folder's name deletes every file synced under it, its subfolders included, for everyone; other folders stay. The dialog names the folder and how many files go, and you type the folder's path to confirm. It needs permission to delete any member's context sources and is not offered while a repository sync is connected. Push the folder again to bring the files back
  • Organize: Create folders for organization
  • Search: Find specific documents quickly

Sync Living Memory from a Repository

Living Memory can read selected folders and files from one branch of a connected repository. In the Living Memory section of the Context tab, choose Sync from repository. The Sync Living Memory from a repository dialog asks for:

  • Repository and Branch;
  • what to sync. Tick folders and files in the tree, or type one under Folder or file and choose Add. You can select up to 50. A ticked folder includes everything inside it, including files added to it later;
  • what to skip. Under Leave out a folder or file, type something inside a selected folder and choose Leave out.

The tree works the same way as the one for coding instructions. Above it you can search, choose Show only selected, or use Collapse all, Select all and Select none. It is a single stop when you press Tab: the arrow keys, Home and End move, Enter opens or closes a folder, Space ticks or unticks a row, and typing the start of a name jumps to it. A partly left-out folder shows a dash, and a row you cannot tick says why. If the repository type cannot be browsed, or the branch is too large to list in full, type the paths instead.

Before you save, the dialog tells you that:

  • each selected folder's own .contextignore is read and applied the same way fabric context push applies it;
  • a file that is no longer selected, is left out, or was removed from the repository is pruned from Living Memory on the next sync;
  • every later sync runs as you, the member who configured it.

Select Keep in sync automatically to pick up changes without choosing Sync now. Fabric picks up a change right after each push to a GitHub repository, and otherwise normally 15 to 20 minutes after a push. Choose Save and sync to start the first sync.

A file the sync cannot store, such as one that is not text, is empty, or is too large, is skipped and named in the sync status. To stop one synced file from coming back, choose Remove from Living Memory on it: the file stays in the repository, and you can restore it from the list of left-out paths.

The menu beside Sync now has Change branch or paths…, Write folders and Disconnect. Disconnecting keeps the files the sync brought in as ordinary synced files; delete them from the tab afterwards if you want them gone.

Write Folders

When a project's Living Memory is synced from a repository, you can name write folders: folders in that repository where agents connected through the Fabric plugin save their files, such as specs and task notes. Fabric commits the files an agent writes there to the repository, and the next sync reads them back into Living Memory, where the next agent sees them.

To manage them, open the menu beside Sync now in the Living Memory section and choose Write folders.

  • To add a folder, pick one in the tree, which shows only folders the sync selects, or type its path. Folders the sync leaves out, hidden folders such as .github, and the folder your coding instructions are read from can't be write folders. Folders can't contain one another, and a project can have up to ten.
  • Each folder sets how its files land. Pull request (the default) opens a pull request against the synced branch, so someone reviews the files before they join it. Push directly commits straight to the branch with no review, and the next agent reads those files as they are. If another write lands on the branch first, Fabric retries the commit on the new branch tip; the write becomes a pull request only when the branch refuses the push (branch protection, a server hook, or missing write permission). Only project admins and owners can turn on Push directly, Fabric asks you to confirm, and the list shows who turned it on and when.
  • Each folder also sets whether AI search uses its files. Index for AI search (the default) embeds the files, so project search and context summaries can use them. Store only keeps the files in Living Memory, readable in the Context tab, through the context tools and in the CLI mirror, but never embeds them, so AI search, keyword search and context summaries leave them out. In the Context tab each such file shows as Ready with a Store only label, and Context Readiness counts it as Ready, not as a source still processing. Use Store only for folders that will hold thousands of task or spec files, which would otherwise fill search and summaries with noise. A Store only badge marks such folders in the list. Changing the choice takes effect without a new sync: switching to Store only removes the folder's existing search data, and switching back embeds the files again (a large folder finishes over its next sync).
  • You can change a folder's path, mode or search choice, or remove the folder. Removing a folder stops agents writing there, and its files are indexed again; files already written stay in the repository.

Members who can't add context can open the list but not change it.

Decisions Found in Living Memory

Once a synced file's new version has been indexed, Fabric reads it with AI and offers the decisions it states as suggestions in the project's Decisions tab. Only a file's current version is offered. A statement you already created or dismissed from the same file is not offered again.

  • What is read: Files marked Store only are not read. Files a repository's first sync wrote are skipped too, because reading a whole repository takes one AI request per file. The Decisions tab says how many synced files have not been read for decisions yet, which also counts files synced before Fabric read decisions from Living Memory; choose Scan for decisions to read them all. Nothing is read until you ask.
  • Who is told: When a scan you started finds decisions, the project's other members who can create decisions get a notification, unless they turned project notifications off. You are not notified, because the suggestions arrive in the tab you are looking at.
  • Limits: Fabric reads the first 60,000 characters of a file and keeps the first 50 decisions it finds. When a file meets either limit, the Decisions tab names the file and says which limit applied, so you know to review the rest of it yourself.
  • Linked work items: When a decision names one of the project's features or bugs, such as F-012, the suggestion links to it. If the decision names none, the file's path or text is used, but only when it names exactly one. A decision you create from the suggestion keeps the link in its details. A work item that was deleted is not linked.
  • Nothing is binding until accepted: A suggestion binds nobody. Reject it, or create a decision from it. A created decision starts as proposed, and is not accepted until someone accepts it.
  • Updates: When a suggestion reads like an existing decision that is accepted or proposed, the review offers Create as update as well as Create as new decision. Creating it as an update links the two, and accepting the new decision marks the one it updates as superseded. A statement that only matches a rejected, deprecated or superseded decision is offered as new.

Team Collaboration

Sharing Projects

Share projects with your organization:

  1. Open project settings
  2. Click Sharing
  3. Choose visibility:
    • Private — Only you
    • Organization — All org members
    • Specific members — Selected individuals

Member Roles

RolePermissions
OwnerFull control, can delete project
EditorCan create and edit documents
ViewerCan view documents only

Activity Feed

Track project activity:

  • Document created by [User]
  • Document edited by [User]
  • Library document added
  • Conversation started

Project Templates

Using Templates

Start from a template for faster setup:

  1. Click Create Project
  2. Select Use Template
  3. Choose from available templates
  4. Customize for your needs

Available Templates

  • Feature Development — PRD, spec, stories
  • API Project — API spec, documentation
  • Research Project — Analysis, findings, recommendations
  • Sprint Planning — Stories, tasks, estimates

Creating Templates

Create templates from existing projects:

  1. Open a project
  2. Click Settings → Save as Template
  3. Choose what to include:
    • Document structure
    • Default prompts
    • Library organization
  4. Save the template

Project Settings

General Settings

  • Name and description — Update project details
  • Default agent — Which agent to use by default
  • Default document type — Starting document type

Prompt Settings

Configure project-specific prompts:

  • Override organization prompts — Use custom prompts
  • Variable defaults — Pre-fill template variables
  • Output preferences — Format, length, style

Integration Settings

Connect external tools:

  • Jira — Sync stories and tasks
  • GitHub — Link to repository
  • Slack — Send notifications

Delegating Coding Runs to Fabric Agents

When your project is linked to a GitHub repository, use Launch Fabric Desktop for installation and native launch options, or Delegate to Remote Server to manage registered servers. AI Connections opens Portal provider settings for workflows and chat. Desktop and remote coding sessions use the accounts configured on their own execution host.

Desktop links require an installed app registered for fabricagents://. If the browser reports no matching app, install or update Fabric Desktop from Fabric Agents. The launch panel also links to the installed app's Settings and Sources views. The native execution bridge is opt-in (FABRIC_AGENTS_EXECUTION_BRIDGE_ENABLED=true on both the API and Temporal worker). See the deployment guide for operator cutover and rollback details.

Personal vs. Organization Remote Servers

You can register and switch between multiple Personal Servers and Organization Servers inside Project Settings → Implementation Defaults or directly inside the Send to Fabric Agents dialog when starting any coding run:

Server ScopeVisibility & AccessTypical Setup
Personal (PRIVATE)Visible only to you within the project. Register as many personal machines as you use (e.g., My MacBook Pro (:9100), Linux Devbox (Tailscale)).Runs on your workstation or personal devbox using your own signed-in CLI subscriptions (claude, codex, cursor, droid, kimi, grok).
Organization (TEAM)Shared with all authorized project members. Registered once by an organization admin or project editor.Runs on a shared team VM or container (https://agents.example.com) with team service accounts or shared subscription seats.

To register an existing server:

  1. Open Delegate to Remote Server from the Coding Agents tab, or the server picker in Start work → Send to Fabric Agents.
  2. Start a matching Fabric Agents server with coding protocol 2, a private workspace profile, Git access, and a supported, signed-in coding backend. The Portal API and Temporal worker must be able to reach its execution endpoint.
  3. Configure a dedicated execution service secret of at least 32 characters as BACKGROUND_AGENTS_INTERNAL_SECRET on that host. Enter the same secret in the Portal connection form. The browser/Desktop RPC token (FABRIC_SERVER_TOKEN) is separate and does not authenticate execution requests.
  4. Enter a connection name, its real endpoint URL, and the execution service secret. Choose Personal or Organization visibility, then click Verify and connect. This checks authenticated server readiness; it does not verify subscription credits.
  5. Select the server and choose a published Roadmap feature. Start work uses that saved server selection and validates readiness for the actual assignment before starting it.

Open remote Web UI opens the selected registered server's address. Sign in there using its own browser authentication. The actual execution link and recorded status are available from the assigned feature's coding run; opening the Web UI alone does not dispatch work.

Triggering a Coding Run

Open a published feature in the Roadmap tab and click Start Work → Send to Fabric Agents… (or click Send to Fabric Agents on an individual story card or task row):

  1. Choose Execution Target: Select Fabric Agents Remote Server (executes in an isolated git worktree on your selected Personal or Organization server) or Fabric Agents Desktop (launches directly in your local desktop app via fabricagents://).
  2. Select Target Remote Server: Pick any of your registered Personal or Organization servers from the visual cards.
  3. Select Task Assignment Scope: Implement the entire feature (all open tasks) or assign a single primary task. When you select a task that specifies its own repository or branch, Fabric automatically targets that repository and branch.
  4. Choose Subscription Harness & Quota Fallback: Pick Claude Max, Codex, Cursor, Factory Droid, Kimi, or Grok, and choose whether exhausted subscription credits should failover across your other subscription harnesses (Free) or fall back to an organization API key.
  5. Click Send to Fabric Agents (or Copy MCP Task Prompt / Open in Fabric Agents Desktop).

The Send to Fabric Agents dialog showing Fabric Agents Remote Server selected, a Personal server chosen, a specific primary task assigned, Claude Max subscription harness selected, and free subscription failover enabled.

Viewing the Live Server Web UI Inside the Fabric Portal

Once a coding run starts, click its status badge to open the Coding Run drawer. You can monitor and interact with the run in two ways without leaving the Fabric Portal:

  1. Embedded Live Server Web UI (<iframe />): Click View Live Server Web UI in Portal in the session header (or Delegate to Remote Server → View Web UI in Portal in the project top bar) to embed the real fabric-agent-server Web UI directly inside the portal. You can watch real-time file diffs, worktree logs, and active workspace sessions inline, or click Open in new tab for a full-window view.
  2. Live Conversation Stream & Follow-Up Instructions: Below the 4-stage progress stepper (1. Worktree & session → 2. Implementing task → 3. Pull Request → 4. Completed), Fabric streams the session messages and provides an inline Send to session box so you can send follow-up instructions to the running or completed session.

The Coding Run drawer showing the 4-stage execution stepper, the embedded Live Fabric Agents Remote Server Web UI inside the Fabric Portal, and the inline follow-up instruction box.

Best Practices

Project Organization

One project per feature/initiative:

✅ "User Authentication v2"
✅ "Q1 Mobile App Redesign"
✅ "API Rate Limiting Implementation"

❌ "All my PRDs"
❌ "Random documents"

Clear naming:

✅ "[Team] Feature Name - Phase"
   "[Platform] API Gateway - Phase 1"
   "[Mobile] Push Notifications - MVP"

❌ "New project"
❌ "test"

Document Management

  • Generate related documents in the same project
  • Upload reference materials to the library
  • Keep the library clean—remove outdated files
  • Use consistent naming for documents

Collaboration

  • Set appropriate sharing levels
  • Add descriptions for team understanding
  • Use activity feed to track changes
  • Archive completed projects

Next Steps