agent-desk
The self-hosted mission control for
Claude Code agents.
Kanban task boards, per-agent chat, a stealth browser, BYO LLM providers, and a proactive dispatcher — all running on your machine.
Your business runs on a dozen apps.
You run between them.
Orders in Shopify, numbers in Sheets, updates in Slack, tickets in Linear, replies buried in Gmail. Every day is the same manual relay — copy this, paste that, chase a reply, rebuild the weekly report. Hiring doesn’t scale it. Rigid zaps snap the moment something changes.
So we built AgentDesk — a team of AI agents that actually do the work, across the tools you already use. Here’s what a day runs like.
Give work to a board, not a new hire.
Drop a card — “pull this week’s Shopify orders into Sheets and flag every refund” — and an agent picks it up, plans the steps, works across each app, and moves it to Done. You manage outcomes on a board, the way you’d run a standup.
The recurring work runs itself.
Put the Monday revenue digest, the daily lead sync, and the month-end invoice chase on a schedule. They fire on cron — overnight, on the hour, whenever — and land finished before anyone thinks to ask.
Steer it in plain English.
Every project has its own chat with full context. “Why did the sync skip 12 rows?” “Redo the report without test accounts.” Ask, correct, and redirect your agents mid-task — no config files, no tickets.
A specialist for every job.
Stand up a research agent, an ops agent, a support agent — each with its own instructions, tools, and memory. The right expert owns the right work, exactly how a real team divides and conquers.
Polish the output without leaving the browser.
Open the deliverable — a doc, a CSV, a draft email — in a full editor inside the dashboard. Fix a line, hand control back, and the agent carries on. No downloads, no copy-paste round-trips.
Everything you need to run an agent team
Kanban Task Board
Project-scoped kanban with statuses (todo, assigned, planning, in-progress, review, done), priority levels, tags, subtasks, and full comment threads with @mentions.
Proactive Agent Dispatcher
Adaptive-rate loop that auto-assigns work. Tick rate self-tunes (15s to 90s) based on activity. Per-agent cooldown, LRU fairness, error backoff, hard timeout per turn.
Multi-User, Multi-Agent
Invite teammates with their own logins. Spin up unlimited agents, each with their own identity, role, LLM provider, and persistent memory.
Per-Agent Chat with Project Context
Multiple sessions per agent — each auto-inherits the current project's context. Persistent main session + unlimited isolated side-sessions. Streams via WebSocket with inline tool calls.
Rich File Explorer & Editor
Built-in markdown editor, CSV viewer in tabular format, HTML/CSS/JS renderer, and syntax-highlighted code viewer. Direct agent control from the file panel. Each project is a real directory on disk.
Stealth Browser (Camoufox)
Fingerprint-spoofing Firefox for live web browsing. REST API: /tabs, /navigate, /snapshot, /click, /fill, /eval, /screenshot. Falls back to Chrome DevTools for authenticated sessions.
Pause / Resume Control
Granular control over projects, individual tasks, scheduled jobs, and the global dispatcher. Pause anything without losing state — resume right where you left off.
Integrations
Telegram notifications + reply from phone. Notion read/write via MCP. Connect your stack — more connectors coming soon.
Scheduled Jobs & Cron
Set up recurring heartbeats, one-shot timers, or cron-expression schedules per agent. Jobs respect pause/resume and show up in the project's schedule dashboard.
BYO LLM Provider
Built on the Claude Agent SDK, optimized for Claude Opus 4.6. Swap providers per-agent, no restart required.
Coming Soon
First-class connectors shipping soon. Agents will read, write, and react across your stack.
Ready to orchestrate?
Want AgentDesk running for your team? Book a call and we'll get you set up.