AgentDock

AgentDock is a local control panel for Codex, Claude Code, and OpenCode. It turns CLI agents into manageable tasks with a Kanban board, live execution trace, follow-up prompts, and workspace selection.

Overview

AgentDock is designed for people who run several CLI agents at once and want a cleaner way to track what each task is doing. Instead of treating every session as a black box, AgentDock keeps the task visible from start to finish:

• create a task
• choose a provider and workspace
• watch it run
• inspect output and session history
• ask follow-up questions
• move it through the board

Everything runs locally.

Features

Task management

• Create Codex, Claude Code, or OpenCode tasks from the dashboard
• Choose a local workspace directory before launching a task
• Set title, description, provider, model, and priority
• Start, complete, cancel, reopen, or resume a task
• Keep task history in local storage

Execution visibility

• Inspect the current command, workspace, provider, model, and session ID
• View runtime status, start time, finish time, and exit code
• Read live output as the agent runs
• Keep a conversation-style history for each follow-up turn
• Render agent responses as Markdown
• Collapse “thinking” text when it is explicitly marked

Kanban workflow

• Organize tasks into Backlog, In Progress, Needs Input, and Complete
• Drag and drop tasks between columns
• See status changes update immediately in the UI
• Keep the board readable with compact cards and clear status badges

Dashboard UX

• Dark glassmorphism interface
• Smooth motion and hover feedback
• Chinese / English UI toggle
• Local-only state refresh with SSE and polling fallback

Why it exists

AgentDock was built for a specific pain point: when multiple Codex, Claude Code, or OpenCode sessions are running, it is easy to lose track of which one is active, which one is blocked, and which one needs a follow-up. This project makes those sessions feel more like manageable tasks than opaque terminals.

Stack

• Backend: Python
• Frontend: Next.js + React + Tailwind CSS
• Drag and drop: dnd-kit
• Animation: framer-motion

Run

Use the root launcher for both services:

./start.sh

If you prefer manual startup:

python3 server.py
cd frontend && npm run dev

Open the dashboard at http://localhost:3000.

Backend API:

http://127.0.0.1:8765

How it works

• Tasks are persisted in data/state.json
• The frontend talks to the Python backend through /api
• Codex, Claude Code, and OpenCode tasks are launched locally and streamed back into the timeline
• Follow-up prompts resume the previous session instead of starting over

Project Structure

• server.py: Python API server and task runtime
• frontend/: Next.js dashboard
• data/: local state storage

Useful Endpoints

• GET /api/state: full snapshot of tasks and sessions
• GET /api/tasks: task list only
• POST /api/tasks: create a task
• POST /api/tasks/:id/start: queue or start a task
• POST /api/tasks/:id/stop: stop a task
• POST /api/tasks/:id/complete: mark a task complete
• POST /api/tasks/:id/block: mark a task blocked
• POST /api/tasks/:id/restart: restart a task
• POST /api/tasks/:id/prompt: send a follow-up prompt

Notes

• The Python backend no longer serves the UI
• If you change backend code, restart python3 server.py
• If you change frontend code, restart npm run dev
• The repo currently targets local usage only

Roadmap

• More provider integrations
• Better session timeline visualization
• Task templates and presets
• More granular runtime telemetry

License

MIT License. See LICENSE.