Dashboard

Patter ships a built-in web dashboard for monitoring calls in real time. It tracks live calls, latency waterfalls, and per-component cost breakdowns, and is embedded in both the Python and TypeScript SDKs from a single source of truth.

Architecture

The dashboard UI is a Vite + React 18 + TypeScript single-page app living at the repo root in dashboard-app/. It is bundled by vite-plugin-singlefile into one self-contained dist/index.html (~190 KB, JS + CSS inlined) and copied into both SDKs as ui.html.

dashboard-app/                ← single source of truth (dev-only, never published)
├── src/
│   ├── App.tsx               # composes the dashboard
│   ├── components/           # Topbar, MetricsPanel, CallTable, LiveCallPanel,
│   │                         # LatencyPanel, CostPanel, PageHeader, ...
│   ├── hooks/                # useDashboardData, useTranscript (SSE-backed)
│   └── styles/               # tokens.css + dashboard.css
├── vite.config.ts            # singlefile plugin config
└── scripts/sync.mjs          # post-build copy to both SDKs

libraries/python/getpatter/dashboard/ui.html   ← synced artefact (committed)
libraries/typescript/src/dashboard/ui.html      ← synced artefact (committed)

At runtime each SDK loads the file as a string:

Python: importlib.resources.files("getpatter.dashboard").joinpath("ui.html").read_text() — see libraries/python/getpatter/dashboard/ui.py.
TypeScript: readFileSync(join(__dirname, "ui.html"), "utf8") — see libraries/typescript/src/dashboard/ui.ts.

The string is served from GET / by the embedded FastAPI / Express app. End users never touch Node — ui.html is checked into both packages so pip install getpatter and npm install getpatter work out of the box.

Local development

The dashboard talks to a running SDK server over the existing /api/dashboard/* REST + SSE API. The Vite dev server proxies /api to 127.0.0.1:8000 so you can iterate on the UI against a real backend.

# Terminal 1 — start any Patter agent on :8000
python my_agent.py                   # your Patter agent (Python or TS)

# Terminal 2 — Vite dev server with hot reload
cd dashboard-app
npm install
npm run dev                          # opens http://127.0.0.1:5173

npm run dev proxies /api/* and /sse/* to http://127.0.0.1:8000, so the SPA renders against whatever the running SDK reports.

Build and sync into the SDKs

npm run build && npm run sync is the canonical pre-publish step:

cd dashboard-app
npm run build      # tsc --noEmit && vite build → dist/index.html (~190 KB single file)
npm run sync       # copy dist/index.html into both SDKs as ui.html

scripts/sync.mjs copies the bundled file to:

libraries/python/getpatter/dashboard/ui.html (shipped via [tool.setuptools.package-data])
libraries/typescript/src/dashboard/ui.html (copied to dist/dashboard/ui.html by tsup)

Both files are committed to the repo. CI rebuilds and re-syncs on every release tag, so a hand-edited ui.html will be overwritten — always edit dashboard-app/src/.

Embedded dashboard

When running serve(), the dashboard is enabled by default at the root URL:

await phone.serve(agent=agent, port=8000, dashboard=True)
# Dashboard: http://127.0.0.1:8000/

await phone.serve(agent, { port: 8000, dashboard: true });
// Dashboard: http://127.0.0.1:8000/

Standalone dashboard

Run the dashboard as a separate process — useful when using the SDK as a tool (e.g. from LangGraph or Claude Code).

pip install getpatter
getpatter dashboard --port 8000

npx getpatter dashboard --port 8000

The standalone dashboard receives call data automatically from the SDK over HTTP.

Features

Top metrics row — Total Calls, Total Cost, Avg Duration, Avg Latency. Numerics in JetBrains Mono so columns align.
Call table — every completed call with cost, latency, turn count, provider, voice mode.
Live-call right rail — caller/callee, streaming transcript, per-call latency waterfall (STT, LLM first-token, TTS first-byte, total turn), and cost breakdown updating as the call progresses.
Call detail — click any historical call for the full transcript, averaged + P95 latency waterfall, and cost breakdown.
Real-time updates — Server-Sent Events on /api/dashboard/events; no polling.
Design system — peach accent, dot-grid surfaces, JetBrains Mono numerics, matching the rest of the Patter brand.

API endpoints

The SPA is a thin client over a stable REST + SSE API — useful for direct integrations.

Endpoint	Description
`GET /`	Dashboard web UI (the bundled `ui.html`).
`GET /api/dashboard/calls`	List calls (paginated).
`GET /api/dashboard/calls/:id`	Single call detail.
`GET /api/dashboard/active`	Active calls.
`GET /api/dashboard/aggregates`	Aggregate stats.
`GET /api/dashboard/events`	SSE event stream.
`GET /api/dashboard/export/calls`	Export as CSV (`?format=csv`) or JSON.

​Dashboard

​Architecture

​Local development

​Build and sync into the SDKs

​Embedded dashboard

​Standalone dashboard

​Features

​API endpoints