Now in private beta · built and run by Nextlab every day Join →

The only tool you need
to build with agents.

Issues, agents, machines and release trains in one fast place. AgentTask replaces Linear and Superset, and ships your backlog while you sleep.

For builders and teams at any level. Experts love it. Beginners ship with it.

◆
Agent opened PR #214AGT-122 · merged into train in 4m
⚡
Placed on Mac minislot 2 of 3 · starts in 0:42
✓
Gate green1,456 tests · 49s on the fleet
app.agenttask.io/nextlab/team/AGT

AgentTask

Train · insight-14 agents running
AGT-134 · claudemac-mini · tmux

Used by experts and beginners. Built by a studio that ships five products on it.

01 · PLAN

Faster than you think. Literally.

Every view loads from your device, not the server. Issues, projects, initiatives and views open before your finger leaves the key.

⌘K everything. Keyboard first, mouse optional.
Specs, sequence and readiness on every issue, so agents never guess.
Works offline. Syncs the moment you are back.
⌘K
Queue AGT-134 for an agentR
Start release train from projectT
Open terminal on Mac mini⇧T
Assign to meI
Move to In reviewS
QUEUE · 7
FLEET · 3 ONLINE
02 · FLEET

Every machine you own becomes a coworker.

Pair a Mac, a Linux box or a cloud server with one command. Queue an issue and AgentTask places it on the machine with the shortest wait, with the right agent, account and model.

No ports, no VPN. One signed binary, one outbound connection.
Rate limit on one subscription? Work moves to the next, then resumes.
Private machines stay yours. Team machines serve everyone.
03 · RELEASE TRAINS

Ship twenty issues before breakfast.

Hand a project to a train. Agents work in parallel, conflicts resolve themselves, the gate runs on your fleet, and every green PR merges into the release branch. You review one ship PR. That's it.

insight-111 merged · 4 running · 7 queued
ETA 2h 10mauto-merge on
04 · TERMINALS

Your agents, in your pocket.

Every session lives in tmux on your machines and streams to any device. Watch a run from the couch, answer its question from the bus, take over the shell when you feel like it.

Sessions survive closed laptops and dead Wi-Fi.
Agents ask, you answer in one tap.
A mobile keybar that actually knows Ctrl and Esc.
AGT-136 · claude● live
Agent asksOverview chart: stack people and agents, or show them side by side?
Stack themSide by side
EscTabCtrl↑⏎
Subscriptions stretched
0%
same first-try success
Agent share of lines
0%
↑ from 76%
Cost per merged issue
$0
↓ 23% with save mode
PeopleAgents, credited to who started them
05 · INSIGHT

Know exactly what your agents cost and ship.

Tokens per run, quota per subscription, lines per person, human next to agent. Save mode picks the cheapest model that keeps quality, and proves it with a control group.

Who used which share of every AI subscription.
Every agent commit credited to the person who started it.
Unknown is shown as unknown. Never a made-up zero.
06 · EVERYTHING ELSE

All the boring parts, done properly.

Bring your own agents

Claude Code, Codex, Grok. Your logins stay on your machines. We never see them, we never store them.

◆ Claude Code◇ Codex✦ Grok

Integrations that move the board

PRs link and move issues. Slack messages become issues. Sentry errors land on the right team.

GitHubSlackSentryAPI + MCP

Merge policy with a spine

Never merges a red gate, a draft or a conflict. Never touches main. Ever.

Signed everything

Every daemon build is ed25519 signed and verified before it runs. Machines update themselves.

Import from Linear

Same keys, full history, mirrored while you try it. Leave when you're ready, not before.

07 · ONE TOOL

Cancel three subscriptions. Keep the good parts.

Linear for issues
Superset for agents
Scripts that babysit release trains
Spreadsheets for AI spend
→

One fast place where people plan, agents build and trains ship. On your machines, with your agents, at your pace.

Go to bed.
Wake up shipped.

Pair your first machine in under a minute. Free while we're in beta.