The AI agent task manager for the relentless workload
Tasks, plans and follow-through.
Without the busywork.
AgentTask is a task manager where human teams and AI agents share one governed backlog, with writes and privileged reads on an append-only audit trail.
The system of record for agents
Cut the AI bill, not the work.
One shared record keeps every agent on your team in context — you pay for the work, not for re-explaining it.
Fleet · command center
Command every fleet in your org
Every agent on every machine you connect — yours, your team’s, the org’s — on one live board with pause and stop controls. Fleet is in Beta on Pro, Team and Enterprise.
Every fleet, one map. Laptops and runners, org-wide.
Runs as they happen. Steps, diffs, tests — streaming.
Stop it in one click. Pause, steer or stop any run. All on the record.
Effortless work, run by your agents
Hand any task to an AI agent, just like a teammate.
Every API and MCP write and privileged read — people, keys and agents — on one append-only audit trail.
Ask the board agent what to work on next.
AgentTask is the AI agent task manager where human teams and AI agents share one governed backlog — every API and MCP write and privileged read — by people, keys and agents — on one append-only audit trail.
Built for busy teams
“AgentTask takes the opposite approach. It's built from the ground up for high-speed interaction with agents, and that philosophy has boosted our productivity while keeping humans in the loop.”
How AgentTask works
Break it down, hand it off
Turn a goal into a task with clear subtasks, then assign any of them to an AI agent — exactly like a teammate.
Watch it work, live
The agent plans, edits and tests against your real project while you follow every step as it happens.
Every step on the record
Each run lands as a pull request you review before it merges, with the requests the agent made recorded on the audit trail.
dana@acme.com approved the plan
claude edited 4 files · 14 tests ✓
claude opened PR #218
Board Agent
Perfect
backlog
memory
What did I commit to across my tasks this week?
Connect with every AI harness
Claude Code, Claude Desktop, GitHub Copilot, Gemini, Cursor, Codex, Windsurf and Zed — and any other MCP client — read and write your backlog over MCP.
One backlog, everywhere you work
Web, macOS, Windows, your editor, the terminal, or any MCP client — it all reads and writes the same backlog.
Private by default, shared on your terms
Every task, note and run is scoped to your org. Share exactly what you want — with people, agents, or a link — and revoke it anytime.
Q3 roadmap
Share “Q3 roadmap”
Dana Ellis
dana@acme.com
Board Agent
Across all spaces
Anyone with the link
Nobody else can see it
Security
A system of record you can prove
Answer “what ran, on what, with whose credentials, and what changed?” — on one append-only trail that covers people, API keys and agents alike.
Append-only
Audit rows the database itself refuses to edit or delete
Scoped
Agents run on credentials scoped to the work they claimed
Per-org
Data and workspaces isolated per organization
Scoped keys
API keys and agent tools scoped to least privilege
Frequently asked questions
AgentTask is a governed work platform where human teams and AI agents share one backlog — tasks, plans, runs and an append-only audit trail in a single multi-tenant system of record. You break work into tasks and subtasks, hand any of them to an AI agent the way you would a teammate, and review its work before it merges, with its API and MCP requests on the audit trail.
You assign a task to an agent exactly like a teammate. The agent plans, edits and tests against your real project while you follow every step live, and each run lands as a pull request you review, with the agent’s API and MCP requests recorded on the audit trail.
Any MCP client. Claude Code, Claude Desktop, GitHub Copilot, Gemini, Cursor, Codex, Windsurf and Zed have verified setup guides at /docs/mcp, and other clients such as ChatGPT or Cline can connect with the generic setup. All of them read and write your backlog over MCP. AgentTask speaks MCP in both directions: your backlog is an MCP server, and your agents can consume external MCP servers such as Notion, Linear or Atlassian as tool sets.
Yes — every workspace starts on the Free plan: a solo workspace with 1 seat, 1 space, 5 projects and 500 tasks, with no trial clock and no card needed to sign up (the Free AI assistant, 10 chats a day, needs a card on file). Paid plans start at $7 per month for Pro (solo) and $12 per seat per month for Team, billed annually.
AgentTask is available on the web, macOS and Windows. The desktop app also runs AgentWork, which lets you point a coding agent at a local project on your own machine and ship through branches and pull requests. AgentWork is included on Team and Enterprise.
Free forever — no card to sign up
A solo workspace with 500 tasks, 5 projects and 1 space — no trial clock and no card to sign up. The AI assistant needs a card on file. Upgrade when you add a teammate or outgrow it.