A real-time dashboard for spawning, directing, and monitoring AI agents as they tackle research, code, writing, and testing tasks — all in one place.
Built with Next.js 16 App Router, SQLite + Drizzle ORM, Server-Sent Events, and the Claude CLI.
OpenWaypoint gives you a Kanban-style project board where AI agents move cards through columns automatically. Spawn a researcher, hand its output to a coder, let a senior coder review it, and finally have a tester write and run test cases — all without manual hand-offs.
Spawn and monitor agents across five specialisations — researcher, coder, senior-coder, tester, and writer — with live status, token counts, cost, and duration tracked per run.
Organise work into projects. Each project gets a Kanban board (Backlog → Planning → In Progress → Testing → Done). Cards move automatically as agents complete pipeline stages, and the active pipeline stage is visualised on each card in real time.
Switch any project to Analytics view for a time-range breakdown of tasks completed, token usage, cumulative cost, and cost by agent role — all backed by Recharts and a fast SQLite aggregation query.
Full run history across all projects with filtering by role, status, and date range. 157 runs and counting.
Toggle --dangerously-skip-permissions for the Claude CLI on a per-instance basis directly from the UI — no config file edits needed.
researcher → coder → senior-coder → tester
↑ |
└──── review cycle ────────┘ (up to 2×)
- Researcher gathers context and produces a research summary
- Coder implements based on the summary
- Senior-Coder reviews — emits
VERDICT: APPROVEDorVERDICT: NEEDS_REVISION - Tester writes test files, runs the project's test runner, and emits
VERDICT: TESTS PASSEDorVERDICT: TESTS FAILED - On failure the coder retries (max 2 cycles), then the task is marked done
| Layer | Choice |
|---|---|
| Framework | Next.js 15 (App Router) |
| Database | SQLite via Drizzle ORM |
| Real-time | Server-Sent Events (/api/stream) |
| AI backend | Claude CLI (claude binary) |
| Styling | Tailwind CSS + Dracula theme |
| Charts | Recharts |
| Tests | Vitest (unit + integration, 100+ tests) |
| CI | GitHub Actions (ubuntu-latest, Node 22) |
- Node.js 22+
- Claude CLI installed and authenticated
git clone https://github.com/JLRansom/openwaypoint.git
cd openwaypoint
npm install
npm run devOpen http://localhost:3000.
The SQLite database is created automatically on first run at data/openwaypoint.db. Drizzle migrations run at startup — no manual steps needed.
npm run dev # start dev server (localhost:3000)
npm run build # production build
npm run lint # ESLint check
npm run lint:fix # ESLint auto-fix
npm run type-check # tsc --noEmit
npm run test:run # full test suite (Vitest)
npm run test:ui # Vitest UI (browser)
npm run test:coverage # coverage reportSee docs/architecture.md for the full reference.
Key conventions:
- API routes live in
/app/api/[resource]/route.ts - In-memory singleton store in
/lib/store.ts— importgetStore()for all state access - Agent simulation / Claude CLI integration in
/lib/agent-runner.ts - Real-time updates via SSE at
/app/api/stream/route.ts - Shared TypeScript types in
/lib/types.ts— import from here, never redefine
All code changes happen in a git worktree — never on main directly:
git worktree add .claude/worktrees/feat-my-thing feat/my-thing
# make changes, open PR, then clean up:
git worktree remove .claude/worktrees/feat-my-thingapp/
api/ Route handlers (agents, tasks, projects, files, stream…)
dashboard/ Agent dashboard page
projects/ Project list + project board + analytics
history/ Run history page
settings/ App settings page
components/ Pure UI components (no store access)
lib/
db/ Drizzle schema, migrations, repositories
executors/ Claude CLI executor abstraction
services/ agentService — pipeline logic, prompt building
types.ts Shared TypeScript types
constants.ts Role colours, pricing, shared constants
docs/ Architecture + agent type reference
context/ Living project memory (roadmap, decisions, sprint)
__tests__/ Vitest unit + integration tests