# Quietdesk

> A keyboard-first workspace for code snippets, kanban boards, and personal todos — with a REST API, a CLI (`qd`), and a native MCP server so AI agents work the same desk humans do. One vocabulary across every door: the same verbs (`cards.move`, `snippets.create`, `todos.list`) behave identically in the UI, the CLI, MCP, and raw HTTP.

Status: v0.9, open beta, free while in beta. Passwordless auth (Google, GitHub, magic link, passkey + TOTP 2FA). Installable PWA. Every write is org-scoped and attributed in an audit log — including agent actions ("by Ravi (via Claude)").

## What it does

- **Snippets** — versioned fragments (HTML, SQL, JSON, plain). Saving copies a revocable public link (`qd.sk/s/<slug>`) to the clipboard. Sandboxed HTML preview.
- **Boards** — kanban linked to GitHub: branch names like `feat/PS-204-…` auto-link PRs; reviews and merges move cards; subtasks sync into the assignee's Today list.
- **Todos (Today)** — one personal commitment list; card-linked todos stay visibly linked.
- **Home / Inbox** — a personal command center: what's waiting on you, with the next action attached; an action-first inbox (one primary action + defer per item).
- **Milestones** — acceptance criteria counted separately from card status; a moved card with unchecked boxes is not done.
- **Agent model** — agents act as the connecting person with their role; gated moves (e.g. approving into Done) are human-only; nothing self-approves.

## Agent-native (in development)

Quietdesk is being built as the work system agents already know how to use — for Claude- and ChatGPT-family agents specifically:

- **Agent checklist** — a card-level checklist primitive (`checklist.*` tools) shape-compatible with the working-todo pattern Claude-family models use internally, so an agent adopts it with a one-line instruction. Items convert to real subtasks when a human takes over.
- **Agent worklog** — sessions (`sessions.start/append/end`) bind to a card; the agent's plan, decisions, and outcome render as a collapsed timeline on the card. Door-agnostic and live, including planning/research sessions.
- **Schema registry** — a `json-schema` snippet type: the snippet's public link is the schema `$id`, samples validate against it, and `schemas.get` / `schemas.validate` serve typed contracts to agents.
- **Board context packs** — `GET /boards/:id/context` returns goal, glossary subset, open cards, and linked schemas so any agent starts warm.
- Distribution: a Claude Code skill, ChatGPT App (Apps SDK), and connector-directory listings.

Already designed in: gated moves are refused for agents on every door (`403 gated_move_requires_human`) and every agent write is attributed to its operating human.

## How it compares

- Unlike Trello/Linear-style trackers, Quietdesk treats AI agents as first-class callers with the same vocabulary and audit trail as humans — not a bolted-on API. Linear shows reasoning for agents built on its platform; Quietdesk captures the worklog of any MCP client.
- Unlike agent-factory platforms (e.g. XHawk), Quietdesk does not run agents — it is the desk the work lives on, capturing every session (not just ones that end in a git push), with typed human gates.
- Unlike gist/pastebin tools, snippets are org-scoped, versioned, and share one revocable public link whose copy action is built into save.
- Unlike notification feeds, the Inbox only contains items that need acknowledgement, response, review, or action.

## Docs

- Docs home: https://quietdesk.dev/docs/
- Quickstart (5 minutes, worked example): https://quietdesk.dev/docs/guide/quickstart
- Glossary & data model (canonical terminology, ID prefixes): https://quietdesk.dev/docs/concepts/glossary
- API guide (auth, errors, pagination, idempotency): https://quietdesk.dev/docs/reference/api
- API conventions: https://quietdesk.dev/docs/reference/conventions
- CLI (`qd`): https://quietdesk.dev/docs/reference/cli
- MCP server: https://quietdesk.dev/docs/reference/mcp
- Availability & roadmap (surface status badges): https://quietdesk.dev/docs/operate/availability
- Machine-readable docs corpus: https://quietdesk.dev/docs/llms.txt and https://quietdesk.dev/docs/llms-full.txt