ADHDev Agent Dashboard Hub

What does this agent do, and when should you use it?

ADHDev is a self-hosted control plane for AI coding agents. It does not replace your agents or spawn its own: it attaches to the CLIs already installed on your machine (Claude Code, Codex, Cursor Agent, Antigravity CLI, Grok CLI, Kimi, OpenCode via PTY), to IDEs through the Chrome DevTools Protocol (Cursor, VS Code, Antigravity, Kiro, Windsurf and others), and to IDE extensions through CDP webviews. A local daemon serves a dashboard at localhost:3847 over HTTP/WebSocket, while a separate adhdev-sessiond process owns the PTY runtimes so CLI sessions survive daemon restarts or upgrades. The centerpiece is Repo Mesh: tasks enter a SQLite queue (pending → assigned → completed/failed with depends_on ordering and retries), idle nodes claim work themselves, each task runs in its own bootstrapped git worktree, and the Refinery validates, checks patch equivalence, rebases submodule-aware, and merges to main only if main has not moved. Quota-aware routing reads each CLI subscription's 5-hour and weekly windows to pick which slot takes the next task. The open-source edition is AGPL-3.0; the cloud edition adds internet-wide reach, push notifications and project file transfer.

On startup the daemon detects installed agents, launches and drives CLI sessions over PTYs, connects to IDEs over CDP, and exposes both a chat view and a raw terminal view in the web dashboard. You describe a task to a coordinator, which files it via mesh_enqueue_task into the SQLite queue with a difficulty tag; an idle node claims it and runs a worker agent inside a fresh worktree that was bootstrapped first (dependencies installed, native rebuilds, gitignored build outputs). Completions and failures land in an append-only ledger kept in the mesh's local SQLite store, and operating notes learned at runtime are injected into every future coordinator prompt. The Refinery then runs the repo's own validation gates, patch-equivalence detection, a submodule reachability check and atomic pointer bumps, rebases, and merges only when main is unchanged — anything undecidable is held for a human; it never force-pushes. Quota routing reads each machine's remaining 5-hour and weekly subscription windows, skips plans that are nearly exhausted, spends quota that would otherwise expire unused first, and the Tasks tab forecasts which slot takes the next task per difficulty. The dashboard transports chat, commands, screenshots and remote input over an encrypted WebRTC data channel; self-hosted mode also exposes GET /api/v1/status, POST /api/v1/command, /api/v1/runtime/:sessionId/snapshot and events, /api/v1/mux/:workspace/state and control, plus ws://localhost:3847/ws.

1. A developer with a laptop, a desktop and a build box who wants one shared queue where idle machines pull coding tasks on their own.
2. Teams already paying for several CLI subscriptions (Claude Code, Codex, Kimi…) who want unused quota spent before its window resets and cheap models handling routine work.
3. Anyone running parallel agents in a submodule-heavy monorepo who needs unattended rebase, atomic submodule pointer convergence, and test gates before main moves.
4. A developer who wants to read agent output, approve or interrupt work, and send the next instruction from a phone instead of babysitting a terminal.
5. Engineers who want a multi-perspective review — the same RCA or design question sent to 2–3 workers on different providers, with agreements and disagreements laid out.
6. Teams that need an auditable trail of every dispatch, completion, failure and stall for later investigation.
7. Developers with an in-house or niche coding CLI they want to surface in the same dashboard via a provider.v1.json manifest.

How do you install or deploy this agent?

Requirements: Node.js 22 LTS, git, and at least one coding agent already installed and authenticated. On Windows, Node.js 24+ is currently blocked for normal startup, so use 22.x or the PowerShell installer.

Recommended path with the adhdev CLI:

npm install -g adhdev
adhdev standalone

Or install the standalone daemon package directly (both install an adhdev command — install one, not both):

npm install -g @adhdev/daemon-standalone
adhdev-standalone

Cloud edition via install scripts or Homebrew:

curl -fsSL https://adhf.dev/install | sh

irm https://adhf.dev/install.ps1 | iex

brew tap vilmire/adhdev && brew install adhdev
adhdev setup

Building from source requires a submodule-aware clone because vendor/seqscribe is a build dependency:

git clone --recurse-submodules https://github.com/vilmire/adhdev.git
cd adhdev
npm install
npm run build
npm run dev

How do you use this agent?

Open http://localhost:3847 after starting the daemon; the dashboard detects which agents are installed on this machine. Useful flags:

adhdev standalone --host 0.0.0.0   # allow other devices on the same LAN
adhdev standalone --port 8080      # custom port
adhdev standalone --token mysecret # token auth for scripts / operator access
adhdev standalone --no-open        # don't auto-open the browser
adhdev standalone --dev            # enable local DevServer endpoints (:19280)

Start a session by picking a provider and a working directory, then toggle between chat and raw terminal. Permission prompts land in the activity inbox instead of blocking an unwatched terminal. To try the mesh on one machine, open /mesh, create a mesh bound to your repo, clone a worktree node, queue a task and watch the ledger go dispatch → completion → Refinery → rebase and merge into main. A second machine on the same network or Tailscale is paired from the mesh screen with its address and a one-time code.

The local API is callable directly:

curl http://localhost:3847/api/v1/status

curl -X POST http://localhost:3847/api/v1/command -H 'Content-Type: application/json' -d '{"type":"..."}'

Custom provider manifests are dropped here:

~/.adhdev/providers/provider.v1.json

What are this agent's strengths and limitations?

How does this agent compare with similar options?

Key facts side by side with the most closely related agents.

How does FollowAgents rate this agent?

FAQ