Monaco, in the page
Open the file the agent just touched, side by side with the chat. Diff viewer, file tree, real terminal per folder.
Open source · self hosted
Every pi coding agent session mirrored to the browser as it runs. Read the transcript, answer the prompt, kill the runaway, resume it from your phone. No screen sharing, no tmux gymnastics.
Why it exists
A long agent run is mostly waiting, punctuated by one question you need to answer. PI Dashboard turns that into a glanceable list: which session is working, which one stopped, which one is blocked on you.
Sessions group by project folder. Pin the folders you care about, name the runs, and stop guessing which terminal tab holds what.
Open the file the agent just touched, side by side with the chat. Diff viewer, file tree, real terminal per folder.
Spin up a git worktree per OpenSpec change. Switch branch from the card. Review the diff and commit without leaving.
Run agents in parallel and watch every one: agent cards, live graph, budget draw-down. Pop a subagent out and read its full timeline.
Schedule agent sessions with cron, or start one when a file lands. Set a goal with success criteria and let the supervisor drive.
Sign in to Anthropic, Codex, Copilot, Gemini. Swap the model mid-session. Point other tools at the built-in proxy.
Take over from your phone: scan a QR to pair. Share over a zrok tunnel with OAuth and an allowlist. Install it as a PWA.
One POST /mcp endpoint puts every session behind an MCP tool call — Claude
Desktop, Cursor, or another pi. Bearer credential required, localhost included.
Ship procedures as skills the agent loads on the right trigger: review, debug, harden, estimate, cut a release. Browse, install and move them per folder or globally.
The UI is composed of named slots. A plugin claims one with a React component — card badge, content view, settings section, tool renderer — and ships as an npm package.
Download
Pre-built installers for macOS, Linux and Windows. Bundles Node.js and auto-installs pi + OpenSpec — you don't need any prerequisites.
Prefer a CLI install? pi package and npm options below.
Get started
Pick whichever fits your workflow. You can always switch later.
Zero prerequisites. Guided setup wizard.
# Download an installer: # github.com/BlackBeltTechnology/pi-agent-dashboard/releases # # macOS — .dmg (arm64 / x64) # Linux — .deb / .AppImage # Windows — .exe (installer) / .zip
Bundles Node.js, auto-installs pi + openspec. System tray integration. Builds are not yet code-signed — see the unblocking notes below.
Windows: SmartScreen warns on first launch. Either More info → Run anyway,
or right-click the downloaded .exe / .zip → Properties →
tick Unblock → OK before running. For ZIPs, unblock the archive before extracting.
macOS: the DMGs are not yet notarized. Either right-click PI-Dashboard.app → Open the first time, or clear the quarantine attribute:
xattr -cr /Applications/PI-Dashboard.app
If you see No such xattr: com.apple.quarantine from an older
xattr -d command, that's harmless — the attribute was never set. Just launch the app.
Already using pi? Add it as a package.
# From inside any pi session: /packages add @blackbelt-technology/pi-agent-dashboard # That's it. The dashboard auto-starts # next time you launch pi.
Classic Node.js install. Runs everywhere.
npm install -g @blackbelt-technology/pi-agent-dashboard # Start the server: pi-dashboard start # or in dev mode: pi-dashboard start --dev
Requires Node.js 20+ and a working pi installation.
FAQ
No. PI Dashboard works only with the pi coding agent; Oh My Pi isn't supported either. Your models still work through pi: sign in to Anthropic, Codex, Copilot or Gemini and pick the model per session.
It depends on the route. The desktop app bundles Node.js and installs pi for you. The pi package route runs inside pi, so pi is already there. The npm route needs Node.js 20+ and a working pi installation. All three are under Get started.
It is open source under the MIT licence and self-hosted: the dashboard runs on your own machine, next to your pi sessions.
Yes. Scan a QR code to pair your phone, or share the dashboard over a zrok tunnel protected by OAuth and an allowlist. You can install it as a PWA.
Sign in to Anthropic, Codex, Copilot and Gemini, and swap the model mid-session. You can also add OpenAI-compatible, Anthropic-compatible or Google Generative AI endpoints as custom providers.
The builds are not yet code-signed. On Windows, SmartScreen warns on first launch: choose
More info → Run anyway. On macOS, right-click the app and choose Open, or run
xattr -cr /Applications/PI-Dashboard.app. Full steps are in the
install notes.