A local-first control plane for AI coding agents.
Flockctl is a daemon that lets you plan work, run coding tasks on your own machine with Claude Code, track cost and status, and manage everything from a web UI or API — without Docker, external databases, or a cloud account.
┌─────────────────────────────┐
│ Web UI ─ CLI ─ API │
└──────────────┬──────────────┘
│ HTTP + SSE + WebSocket
┌──────────────▼──────────────┐
│ Flockctl daemon │
│ scheduler · auto-executor │
│ planner · permission gate │
│ AI client · cost tracker │
│ │
│ SQLite (WAL) │
└─────────────────────────────┘
- Local-first by default. Binds to
127.0.0.1only; your code, API keys, and task history never leave your machine unless you opt in. - One install, zero infra.
npm install+flockctl start. SQLite in WAL mode handles persistence; no Docker, Redis, Postgres, or queue broker. - Plan → slice → execute. Break projects into milestones and slices, let the auto-executor fan out independent work in parallel, gate risky tool calls behind approval.
- Claude-first today. Runs Anthropic Claude via the Claude Code CLI / agent SDK. Multiple keys are supported with priority + automatic failover on rate limits. Other providers (OpenAI, Gemini) are on the roadmap but not wired up yet.
- Cost visible by default. Every session records tokens and USD; the UI breaks usage down by project, model, and key.
- Multi-machine when you want it. Opt into labeled bearer-token auth to reach the same daemon from a phone, laptop, or CI runner over VPN / tailnet / reverse proxy.
| Capability | What it means |
|---|---|
| Workspaces | Top-level grouping for related projects; share AI-key scoping across a set of projects |
| Projects | Git-backed path with per-project model, key scoping, and permission-mode settings |
| Planning | Milestones → slices → tasks hierarchy; LLM-generated or hand-edited |
| Tasks | Prompts routed to Claude (CLI or agent SDK) with live log streaming and approval gates |
| Chats | Multi-turn sessions with tool approval, streamed over SSE, resumable |
| Auto-execution | Dependency-aware scheduler that runs independent slices in parallel, stops on failure, resumes after restart |
| Templates & schedules | Reusable task recipes; cron-driven recurring runs |
| Skills & MCP | Global / workspace / project skill files and MCP server config, merged transparently at runtime |
| Metrics | Cost, duration, and throughput dashboards per project / model / key |
# From npm (release candidate)
npm install -g flockctl@next
flockctl start # http://127.0.0.1:52077# From source
git clone <repo-url> && cd flockctl
npm install && npm run build
npm link # puts `flockctl` on your PATH
flockctl start # http://127.0.0.1:52077Pre-1.0 releases are published under the
nextdist-tag.npm install -g flockctl(without@next) will start resolving once the first stable1.xis cut.
See docs/INSTALLATION.md for prerequisites, platform notes, and the full source-install flow.
A 10-minute walkthrough — create a project, add an AI key, run your first task — lives at docs/GETTING-STARTED.md.
| Start here | |
|---|---|
| Getting started | First-run walkthrough |
| Concepts | Glossary: project, slice, chat, workspace, skill, … |
| Installation | npm + source + prerequisites |
| CLI reference | Every command and flag |
| Configuration | Env vars and ~/.flockctlrc |
| Remote access | Multi-device tokens, reverse proxy, TLS |
| Reference | |
| API | REST, SSE, WebSocket |
| Architecture | Internals and data flow |
| Database | Drizzle schema |
| Protocol | Real-time event shapes |
| Security | Threat model |
| Deployment | Daemon ops, backup, health |
| Development | Contributing and conventions |
| Testing | Test ladder and CI |
See CHANGELOG.md for the list of notable changes per release. The format follows Keep a Changelog and versions follow SemVer.
See LICENSE.



