Repository navigation
Home
One place to see, steer, approve and pay for every AI coding agent you run.
Nautelle (formerly AgentDock) is a self-hosted control plane for Claude Code agent fleets. You describe work in a sentence. Nautelle interviews you and turns the answer into cards on a board. It can draw a design first. Then it runs a dedicated agent for each card on your own machines, and that agent ships a pull request. Everything the agents do, and everything it costs, is in one web app, in Telegram and in the terminal you already use.
Tip
New here? Start with Getting started, then read How work flows. These two pages explain everything else.
| 🚀 Getting started | Run Nautelle locally, pair a runner and connect your first project in about 15 minutes. |
| 🔁 How work flows | The life of a piece of work, from one sentence to a merged pull request. |
| 🧩 Architecture | The control plane, the runner, and what runs where. |
| Board and cards | Columns, cards, dependencies, Auto-run, bulk actions and the Work page. |
| New work intake | Turn one sentence into a card, a split or a stack. Discuss mode is covered here too. |
| Design and stacks | Mockups from your UI profile, reviewed with pins and approved. Staged stacks share one design. |
| Card workers and shipping | Run, rounds and feedback, draft PRs, CI fix rounds, auto-merge and Accept. |
| Fleet and orchestrator | Code Sentinel slots, the queue, approvals, the live pane and terminal attach. |
| Token Saver | Filtered command output, read guards, early compaction and measured savings. |
| Steward | Detects stuck agent work, applies safe fixes and escalates the rest. |
| Context handoff | A short note per card, so every new round starts in context. |
| Sessions, usage and budgets | Session trees, API-equivalent cost, usage breakdowns and budgets. |
| Claude Code integration | The plugin, MCP tools, terminal work on the board and away mode. |
| Telegram | A bot with commands, buttons, voice notes and a Mini App. |
| AI assistant | A floating chat backed by your own LLM provider keys. |
| Automation | Skills, schedules, webhooks and the GitHub App. |
| Installing the runner | Homebrew, npm, pairing, the service and the local security policy. |
| Self-hosting | The reference deployment and its environment variables. |
| Security and access | Accounts, roles, audit, the runner trust model and what leaves your machines. |
| Project settings | Every per-project switch, coding rules included. |
| Contributing | Repository layout, commands, conventions and CI. |
| Glossary · FAQ | Terms and short answers. |
Nautelle is under active development and ships from develop several times a
day. Orchestrator slots and card workers run Claude Code. Codex sessions
are listed, but Codex workers are the next milestone
(roadmap).
The engineering source of truth is
docs/ in the
repository: architecture decision records (ADRs), specifications and the
architecture. Wiki pages link to the decisions they describe.
Nautelle · Website · Discord · Engineering docs · MIT
Start
Work
Cost and health
Surfaces
Operate
Reference