DocOps

• Why DocOps
• Features
• Screenshots
• Quick start
• Development
• Configuration
• How it works
• Contributing
• License

Coding agents are great at focused tasks, but juggling several of them means juggling terminals, branches, half-finished diffs and "wait, which session was that?". DocOps turns every piece of work into a card: a card is a thread, a branch and an isolated worktree. Agents run in parallel, report progress on the board, ask you questions when a decision is genuinely yours, and land their work in a review queue with a running preview and a test plan.

• 🗂️ Kanban board per project — Backlog → In Progress → Human Review → Done → Merged, with drag & drop, colorful labels, checklists, due dates, custom boards and live agent activity on every card.
• 🌿 Thread = branch = worktree — every task gets its own branch and worktree inside the project (<project>/.docops/worktrees/). Follow-up messages resume the same CLI session in the same workspace.
• 🤖 Bring your own agent — run Claude Code, Codex CLI or OpenCode directly, or let the Hermes agent orchestrate while you still choose the CLI, model and reasoning effort per task.
• 🧑‍✈️ Manager agents — hand over your notes (typed, pasted or a photo) and a manager writes the cards, runs them in parallel, verifies the results with your test suite, and reports back.
• 🔌 Built-in MCP server — agents drive the board through kanban tools (get_card, move_card, check_item, ask_user, …) instead of fragile text markers.
• 🙋 Agents can ask you — a card parks and glows when an agent needs a decision; your answer resumes the exact same thread.
• ✅ Review queue — finished work arrives with a running preview, the diff and a generated test plan. Ship it or send it back in one keystroke.
• 💻 VS Code page — changed files, commit box, commit graph, Monaco diff, a real terminal and the agent thread side by side. Alt+C flips to the running app.
• 🔀 One-click AI merge — conflict preview via git merge-tree; clean branches merge instantly, conflicted ones can be resolved by you or delegated to any agent.
• 🚀 Workspace slots & previews — a pool of warm worktrees with one-click app previews (commands or Docker Compose), shared services and live logs.
• 💸 Token and cost visibility — per-task and per-batch usage, split across Hermes, Claude and Codex.
• 🧰 And more — GitHub issue import, agent profiles and pipeline templates, ideation chat → task, animated changelog, embedded terminal and browser-based CLI logins.

Task page — thread, checklist, labels and live agent logs	VS Code page — diff, terminal and agent thread on one screen
Agent profiles — engineering team and BMAD method templates

• Docker with Compose
• A folder containing the git repositories you want agents to work on
• An account for at least one coding CLI (Claude Code, Codex or OpenCode)

git clone https://github.com/Oak-AN/DocOps.git
cd DocOps

Create a .env file next to docker-compose.yml:

PROJECTS_DIR=/home/you/code
KANBAN_HOST_PATH_MAP=/home/you/code=>/projects
KANBAN_SUPERUSER_USERNAME=admin
KANBAN_SUPERUSER_PASSWORD=change-me-to-something-long

On Windows use paths such as PROJECTS_DIR=D:\Code and KANBAN_HOST_PATH_MAP=D:\Code=>/projects.

python local_build_run.py
# or
docker compose up -d --build

Open http://localhost:40000, sign in to your coding CLIs on the Connections page, add a project, and create your first card.

docker compose -f docker-composeDEV.yaml up -d --build

• App (Vite dev server): http://localhost:5001
• Django admin: http://localhost:8002/admin

The code is bind-mounted, so saving a file on the host reloads the app.

cd backend
python -m venv .venv && source .venv/bin/activate
pip install -r requirements.txt
python manage.py migrate
python -m uvicorn backend.asgi:application --port 8000

cd frontend
bun install
bun run dev

Then open http://localhost:5173.

cd backend
python manage.py test api

All state lives in ./runtime next to the compose file (override with RUNTIME_DIR):

Security note: DocOps runs coding agents with access to your repositories and credentials. Run it on a trusted machine or private network, change the default admin password, and set KANBAN_SECRET_KEY and KANBAN_DEBUG=0 before exposing it anywhere.

┌───────────────────────────┐     docker exec       ┌────────────────────────┐
│  DocOps app container     │ ────────────────────▶ │  Hermes agent sidecar  │
│  Django + ninja API       │ ◀──────────────────── │  (nousresearch image)  │
│  React UI (Vite build)    │     CLI proxies       └────────────────────────┘
│  Claude/Codex/OpenCode    │
│  kanban MCP server        │ ◀── agents call board tools (move_card, ask_user, …)
└────────────┬──────────────┘
             │ /projects
             ▼
   your repos + .docops/worktrees/slot-*, t<id>-*

• Backend: Django, django-ninja, SQLite, uvicorn, FastMCP
• Frontend: React, Vite, Tailwind, shadcn/ui, Zustand, Monaco, xterm.js
• Runtime: a lightweight app image beside the official Hermes agent image

Design notes live in .docops/ARCHITECTURE.md and .docops/FEATURES.md.

Contributions are welcome!

1. Fork the repository and create a branch: git checkout -b feat/my-feature
2. Make your change and run the backend tests and bun run lint
3. Commit using Conventional Commits (feat:, fix:, …)
4. Open a pull request describing what changed and why

Found a bug or have an idea? Open an issue.

Released under the MIT License.