Web Dashboard
The web UI is the visual side of Shep — the same features as the CLI and shep tui, plus a canvas of every repo, live logs, diffs, chat and reviews.
Starting it
shep # start Shep in the background and open the web UI
shep status # is it running, and on which port?
shep stop # stop itIt runs at localhost:4050 by default; if that port is busy, Shep picks the next free one and prints the URL. To choose the port:
shep start --port 8080
shep restart --port 8080 # if it's already runningshep ui is different: it runs the web server in the foreground in your current terminal instead of as a background daemon — handy for watching its output. It takes --port and --no-open (don’t open the browser). Stop it with Ctrl+C.
Layout
The sidebar on the left holds the main pages. Collapse it with the toggle at the top or Cmd/Ctrl + B. Below the links, a Features list groups your features by status (action needed, in progress, and so on).
| Page | What it’s for |
|---|---|
| Control Center | The home page: a canvas of your repos and their features, the session tree and the fleet bar. |
| Chat | Conversations with Shep and with agents, global or per repo, feature or app. |
| Agent activity | Every feature’s agent state and why. The badge counts features that need you. |
| Apps | Apps you created with Shep, and the New app flow. |
| Inventory | A searchable table of every app, repository and feature. |
| Project Memory | Durable knowledge agents read at the start of every feature. |
| Spaces, Connections, Opportunities, Incidents, Factory | The software factory — each behind its own feature flag, on by default. |
| Tools | Developer tools Shep can find or install (agent CLIs, gh, editors, terminals). |
| Projects | Work items, cycles and intake (behind the projects flag, on by default). |
| Skills | Skills Shep can inject into a feature’s worktree. |
| Collaboration | Agent Questions, Supervisor and Agents — see Supervisor & guardrails. |
| Plugins, Webhooks | MCP/CLI plugins and GitHub webhooks. |
| Settings | Everything configurable — see below. |
Some pages appear only when their feature flag is on: the SDLC board, Clusters, Learning, Security (ASPM), Workflows and Harness are all off by default.
You can choose which page opens first in Settings → Home page: Control Center, Apps or Features.
Control Center
The Control Center is a canvas: each repository is a node, with its features branching off it. A feature node shows its phase, mode and agent; click it to open the feature drawer.

Session tree. Beside the canvas, the Sessions panel lists Repository → features → agent sessions. It also shows Claude Code, Cursor and Gemini sessions you started yourself in that repo: sessions that became Shep features are marked as adopted, the rest you can archive, restore or delete. To turn an existing branch into a Shep feature, use Adopt Branch from the + button or shep feat adopt <branch>.
Fleet bar. The header shows the state of the whole fleet: how many features are cruising, queued, need you and failed, whether the circuit breaker tripped, and — if the queue is paused — a Resume button. Click it to open the triage drawer with everything that needs a decision. See Fleet management.
Repository node. Buttons on each repo open it in your IDE, a shell or your file manager, start its local dev server (see Dev servers), open a chat about it, or add a feature with + New.
The + button (bottom corner) creates things: New Feature, Local Folder, Folder of repos, Adopt Branch, From GitHub, or Start from a prompt for a new app.
Creating a feature
- Click + New on a repository, or New Feature from the + button. (Or just ask in Chat.)
- Describe what you want built.
- Pick the agent and model, the reasoning effort, and Fast or Spec mode.
- Choose what to automate: Approve PRD / Plan / Merge, Evidence, Skills, and Git (Push, PR, Watch CI, Sync, Commit Specs, Fork & PR).
- Optionally pick a parent feature, attach files, or switch on Pending to create it without starting.
- Click Create Feature (or
Cmd/Ctrl + Enter).

The feature drawer
Click any feature — on the canvas, in the session tree, in the sidebar’s Features list, or on Agent activity — to open its drawer. The top bar has buttons to open the worktree in your IDE, a shell or your file manager, copy the feature ID, start its dev server, archive or delete it. The tabs:
| Tab | What’s there |
|---|---|
| Overview | Description, branch, pinned agent and model, mode, the gate and git settings it runs with — and an Agent status card: why it’s in its current state, what it’s asking, and what comes next. Notify me alerts you when it needs you. |
| Activity | Each phase with its run time, wait time, model and cost, plus the decisions made along the way. |
| Log | The agent’s output, live — structured or raw. |
| Plan | The spec-driven plan and each task’s progress. |
| PRD Review | The requirements questionnaire: each question with options and a recommended answer. Number keys 1–9 pick an option. |
| Tech Decisions / Product | Technical and product decisions from research. |
| Merge Review | The diff, PR and CI status at the merge gate. Approve here and Shep merges. |
| Chat | A live chat with an agent about this feature, in its worktree. |
| Checkpoints | A snapshot after every step: see what each step changed, compare two, or restore one. |
Tabs appear when they apply: spec tabs only for spec-driven features, Prototype for exploration features, Bedrock when Bedrock memory is on, Context when the experimental harness is on.

Approving, rejecting and stopping
When a feature waits at a gate, the drawer’s action bar shows Approve and Reject. Type feedback in the box and press Cmd/Ctrl + Enter to reject with it, or Cmd/Ctrl + Shift + Enter to approve. Stop ends the agent’s current run and leaves the worktree intact; Resume picks it up again; a failed run shows Retry.
The same actions are available as shep feat approve, shep feat reject --reason, shep ps stop, shep feat resume, and as keys in shep tui.
Chat
The Chat page lists your conversations in groups: Shep (a global chat), Repositories, Features and Applications. Agents answer with inline blocks you can act on — a spec to approve or request changes on, an HTML preview, a plan to implement, or a decision with a recommended option.

Cmd/Ctrl + Shift + K opens a chat popup on any page; Cmd/Ctrl + Shift + M maximizes it. Interactive chat runs on Claude Code or Cursor CLI. See Chat.
Agent activity
The Agent activity page answers “what is every agent doing, and why?” — the same answer as shep feat status, MCP and A2A. Features are grouped as Needs you, Working, Waiting, Failed and Done, each with its reason and a link to the next step (review and approve, open to resume, …). Turn on Show finished to include completed ones.

Apps
The Apps page lists the apps you’ve created with Shep. New app offers Plan it first (spec-driven, any stack), Quick prototype (Vite + React + Tailwind + shadcn with a live preview), Open local project and Import from GitHub. See Start an app.
Inventory
Inventory lists every app, repository and feature in one table. Search, filter by status (Action Needed, In Progress, Blocked…), group by repository, and show archived features. Its toolbar can also add a feature, a local repo, an app or a GitHub repo.
Settings
Settings is one long page with a section menu at the top:
| Section | What you set there |
|---|---|
| Language | UI language. |
| Agent | Default agent and model, reasoning effort, auth method and token. |
| Adaptive models | Route simpler tasks to cheaper models in the same family. |
| Environment | Default editor, shell and terminal. |
| Workflow | Default feature mode, max parallel features, question deadline, auto-approve gates, evidence, push / PR, auto-archive. |
| Worktree | Custom create and post-create commands. |
| CI | Max fix attempts, watch timeout, log size, poll interval. |
| Timeouts | Per-stage time limits for the agent. |
| Notifications | Which agent and pull-request events notify you. |
| Messaging | WhatsApp task dispatch (when its flag is on). |
| MCP | The shep mcp command and the tools it exposes. |
| Connect agents | The A2A agent card and token for other agents. |
| Flags | Feature flags — also a full page at /settings/feature-flags. |
| Chat | Chat tab, idle timeout, max concurrent sessions. |
| Home page | Which page opens first. |
| Layout | Position of the floating buttons. |
| Integrations | GitHub token for terminal sessions started from Shep. |
| Usage metrics | Turn anonymous usage metrics on or off. |
| Database | Where the database lives and how big it is. |
Security, Agent Harness and Self-learning sections appear only when their flags are on.
Many of these are also available as shep settings commands. CI, timeouts, default feature mode, question deadline and home page are set only here. See Configuration.
Keyboard shortcuts
| Shortcut | Action |
|---|---|
Cmd/Ctrl + K | Global search |
Cmd/Ctrl + Shift + K | Open or close the chat popup |
Cmd/Ctrl + Shift + M | Maximize or restore the chat popup |
Cmd/Ctrl + B | Collapse or expand the sidebar |
Cmd/Ctrl + Enter | Create the feature (New Feature drawer); reject with your feedback (action bar) |
Cmd/Ctrl + Shift + Enter | Approve (action bar feedback box) |
1–9 | Pick an option in a decision or PRD question |
Esc | Close a drawer or dialog |
Optional: the SDLC board
A Kanban-style board of features by lifecycle phase exists but is off by default. Turn it on in Settings → Flags or with shep settings flags enable sdlcBoard; it then appears in the sidebar.
When to use the web UI vs the terminal
- Web UI for diffs and reviews, chat with inline specs and previews, and a visual map of many features across repos.
shep tuito watch and approve everything from the keyboard, or inside tmux.- The CLI for scripting and quick one-offs.
They all work on the same features — anything you start in one shows up in the others.
What’s next
- Chat — start and steer features in a conversation.
- Fleet management — the fleet bar, triage drawer and queue.
- Git checkpoints — the Checkpoints tab in depth.
- Configuration — what each setting does and where it’s stored.