A personal command-center dashboard for orchestrating AI agents, projects, hardware builds, spend, and day-to-day ops — all from a single pane of glass.
Built with React 18 + TypeScript + Tailwind CSS on the frontend and an Express 5 API server on the backend.
Press ⌘K (Ctrl+K) anywhere for the command palette — jump to any page or search notes, docs, and tasks. Every view runs on real data from your connected agents; views lazy-load on demand.
The sidebar is grouped into four sections — Work, Knowledge, Build, and AI Ops — plus Settings. Related monitoring views are consolidated into tabbed hubs rather than scattered across many sidebar entries.
| View | What you can do |
|---|---|
| Home | Landing overview — mission-control hero, an "at a glance" priority queue (alerts, overdue to-dos, pending approvals, system errors), and quick-view cards for usage, to-dos, projects, alerts, and system health. Every card deep-links to the right hub and tab. |
| To-Do | Personal quick-capture list plus a built-in view switcher — To-Do (default), Tasks (kanban board), Approvals, and Inbox — so personal items and agent work live on one page. To-Do: natural-language quick add (!crit @long tomorrow), severity levels (low → critical), short/long horizon, due dates with overdue badges, inline editing, a one-click OpenClaw/Hermes research button (summary, action steps, links, key facts), an optional Additional Details sub-panel (date, time, location, phone, cost, URL, contact, category, custom fields) with smart auto-detection, and opt-in Google Calendar sync for dated items. Tasks is a status-column kanban (Active / Queue / Blocked / Completed); Approvals reviews agent output (approve / reject / request changes); Inbox is a unified priority-sorted feed of approvals, tasks, and to-dos with snooze. |
| To-Buy | Personal shopping list with the same click-to-open drawer as To-Do. Quick add with priority/quantity/price tokens (power drill !high x1 $89), a running estimated total, and one-click OpenClaw/Hermes research that returns general info, a fair price + range, online buy links, local store options, and key specs — auto-filling the item's price estimate. |
| Spend | Personal money command center. Separates Claude Code (a subscription — shown as token-equivalent value, not billed per-token) from OpenClaw/Hermes agents (real per-token API spend), plus a "things" lane (open To-Buy total + the value of hardware you already own). Pure aggregation over existing endpoints; 30-day trend and projected monthly. |
| Chats | Browse Claude / OpenClaw / Hermes conversation sessions. See token counts, cost per session, and open the full message history for any conversation. |
| Calendar | Day / Week / Month / Agenda views of your scheduled tasks and Google Calendar events, with prev / today / next navigation and an in-app event composer (create / edit / delete). |
| View | What you can do |
|---|---|
| Docs | Documentation browser connected to your agent knowledge base, with a Notes tab (quick-capture scratchpad) and a Links tab — save, tag, pin, and archive bookmarks; add manually or via the Inbox "save as link" action. |
| News | Real-time curated news dashboard with three tabs: Feed (17 RSS/Atom sources across AI, computing, code, and robotics), GitHub (trending repos by time range and language), and Buzz (top discussions from HN, Reddit, and Lobsters). ADHD-friendly magazine-style hero, cross-source "Top right now" strip, category/platform filters, and 5-minute auto-refresh. |
| Memory | A live operational dashboard over your agent's real memory system, read over SSH from the agent machine: an Activity timeline of memory events (recall / dream / promotion, each expandable to the actual content), Daily logs + workspace files with a full-text index over every day, Recall (semantic chunks + similarity scores + concept tags), Dreaming (the nightly Light / Deep / REM consolidation and promotions into long-term MEMORY.md), Health (LanceDB vector store + pipeline freshness/staleness), and Metrics (per-day volume, dream phases, top concepts). Configure the SSH connection via OPENCLAW_SSH_*. |
The hardware → ideas → projects loop in one place.
| View | What you can do |
|---|---|
| Projects | Kanban-style boards for tracking initiatives (drag cards across columns, set owners, attach notes), with a Pipeline tab — a multi-stage run monitor (card grid or Gantt timeline) with step-level traces and a cron jobs panel. |
| Inventory | Hardware catalog with SQLite-backed persistence. Search and filter items, view status (available / in-use / reserved), trigger per-item agent research, and inline-edit any field. Bulk add lets you paste a free-form list (bullets, xN quantities, (Model: …)) — it parses each line into an item, shows a live preview, and can queue agent research on the whole batch so specs, value, and form factor get filled in automatically. |
| Ideas | Idea Factory — browse and triage the buildable project ideas the agents generate from your inventory. Confidence/coolness score bars, difficulty, cost/time, and parts-on-hand / still-need lists. A Buildable now filter surfaces projects you can build with parts you already own; sort by Best / Coolest / Cheapest / Newest, search across ideas, like / snooze / Built it! / reject-with-reason, and trigger a new generation run. |
Five consolidated hubs covering everything your agents do — built on local Claude Code logs plus the OpenClaw (WS) and Hermes (REST) gateways.
| View | What you can do |
|---|---|
| Activity | Live cross-platform agent monitoring in one place: Live (real-time tool calls / file reads / commands streamed from OpenClaw + Hermes), Sessions (run history with token/message totals and full transcripts), Brain (raw event log with tool-loop detection and top tools), Agents (who's running and their current state), and Map (node-link traffic topology, 1 h / 24 h / 7 d / all). |
| Usage | Cost & model analytics: a Claude Code token/cost view (daily trends, token mix, cost anomalies, per-model share, hour-of-day heatmap) plus a Models tab (Helicone-style spend, latency, volume, and failure rate per model with a cost-vs-latency scatter). |
| Benchmarks | Benchmark how a model performs through OpenClaw/Hermes (App → harness → model → tools/context/routing → result), not raw model APIs. 9 behaviour lanes, 4 task packs, deterministic scoring with normalized failure types, multi-sample reliability scoring, a per-task detail drawer, and a model/provider Compare fingerprint (reliability, speed ± stdev, verbosity/tokens, est. cost, fences). Runs real dispatches via the Hermes API server, OpenClaw WS, or any OpenAI-compatible /v1 endpoint (LM Studio / Ollama / vLLM). |
| Evals | Agent performance evaluation hub. Model scorecard leaderboard, agent-model matrix, session trend chart, benchmark task runner (dispatched to the Hermes API server), memory benchmark panel (recall / multihop / temporal / conflict / applied / negative), and a scoring methodology reference. |
| Health | Platform & connector health: System (Claude config / MCP / plugin health, uptime, memory file browser), Security (connector token health, reachability, auth-error counts, risk badge, live diagnostics probes), Alerts (rule builder — error rate / loop / stalled / token spike / no activity — plus fired alerts), and per-platform deep-dive dashboards for OpenClaw and Hermes. |
Configure connectors, API keys, the Google connection (connect / reconnect / disconnect), and application preferences.
- Frontend — React 18, TypeScript, Tailwind CSS, Vite, Lucide icons
- Backend — Express 5, tsx (TypeScript execution), dotenv, Node.js ≥ 18
- Database — SQLite via Node's built-in
node:sqlite(DatabaseSync) for inventory, agent-event, and benchmark stores; JSON files for lighter stores (notes, alerts, connectors) - Integrations — Google Calendar API (OAuth 2), Anthropic API, OpenClaw WebSocket, Hermes REST
- Node.js ≥ 18
- npm ≥ 9
npm installCopy the example env file and fill in your credentials:
cp .env.example .envRequired variables (see .env.example for details):
| Variable | Purpose |
|---|---|
API_PORT |
Express server port (default 3001) |
GOOGLE_CLIENT_ID |
Google OAuth client ID (Calendar) |
GOOGLE_CLIENT_SECRET |
Google OAuth client secret |
ANTHROPIC_API_KEY |
Anthropic API key (Radar analytics) |
GITHUB_TOKEN |
GitHub personal access token — optional; raises the Search API rate limit for the News → GitHub tab |
OPENCLAW_SSH_HOST |
Agent machine host — reads the OpenClaw memory system (daily logs, dreaming, LanceDB recall) over SSH for the Memory view |
OPENCLAW_SSH_USER |
SSH username on the agent machine |
OPENCLAW_SSH_KEY |
SSH private key — absolute path, or a key name resolved under ~/.ssh |
Google Calendar: set the two GOOGLE_* values, then connect from Settings → Google (or visit /api/auth/google). The refresh token is stored in data/google-tokens.json and refreshed automatically — no copy-paste. Publish your OAuth consent screen ("In production") in Google Cloud Console; while it is in "Testing" mode Google expires the refresh token after 7 days.
Start both the API server and Vite dev server:
npm run dev- Frontend →
http://localhost:5173 - API →
http://localhost:3001
npm run buildPrivate — not licensed for redistribution.