The workspace where AI agents actually finish what they start.

TaskTree is the only project management system that connects business planning, development, and deployment into one coherent loop — and lets AI agents drive it end-to-end.

Stop babysitting your agents. Give them a tree, a workflow, and a memory — and let them ship.

Early access
project_graph.tsk
Ship v2 launchPlan releaseBuild featuresDeploy & QAAuth refactorBilling APIWebhooksStripe syncInvoices
agent.session #4f2 · executing

9 nodes · 3 layers · 1 orchestrator autonomous
The problem

AI coding agents are brilliant for an hour and useless for a week.

Drop a real project on them and the wheels come off.

01

They forget yesterday's decisions and redo work.

02

They drift off scope between conversations.

03

They balloon every prompt until the context window collapses.

04

Your business plan, code, and infra live in three different worlds — and the agent only sees one at a time.

You don't need another Kanban board with an AI sidebar. You need structure that AI was built to operate inside.
The solution

A project graph with native workflow primitives that AI agents call as tools.

Planning, preparation, execution, testing, finalization, orchestration. Humans get a clean dashboard. Agents get a structured backbone. The work gets finished.

Plan
tool.plan()
Prepare
tool.prepare()
Execute
tool.execute()
Test
tool.test()
Finalize
tool.finalize()
Orchestrate
tool.orchestrate()
Core features

What's in the tree.

Five primitives that turn a chatbot into a teammate. Each one is a tool the agent calls — not a checkbox a human ticks.

01 · Structure

A tree, not a list.

Goals → projects → tasks → subtasks. Each with three depths of description so the agent zooms in or out without re-reading everything.

planner
# Goal
## Projects
...
executor
task.exec
deps:[..]
spec://
reviewer
spec
criteria
diff
02 · Context

Patterns, not dumps.

Configurable patterns assemble the right slice of the project per role. Planners see high level. Executors see dependencies. Reviewers see the spec.

preparer
executor
tester
52%
finalizer
03 · Roles

Separation of agents.

Different agents prepare, execute, test, and finalize. The reviewer is never the author. Decisions get logged. Resolutions propagate. Quality goes up.

orch
04 · Autonomy

Orchestration that doesn't drift.

An orchestrator drives a whole branch to completion — delegating each step to a fresh subagent — while you sleep. Wake up to a tree of done.

decision use postgres for tasks mar 14
doc auth-spec.md mar 18
resolution rate-limit per user mar 22
decision k8s on fly.io apr 02
dependency billing → invoices apr 05
05 · Memory

A graph that remembers.

Tasks, documents, decisions, and dependencies in one persistent graph. Hand the project to a new agent next month — it picks up where the last left off.

06 · The whole thing

Five primitives.
One coherent loop.

Plan, prepare, execute, test, finalize, orchestrate — wired together by a memory graph that nothing falls out of.

Read the docs
Coming soon

Smart Agent Spawn — start sessions from anywhere.

Kick off agentic sessions from your phone, your laptop, or another agent. Point at a task, say "go." The session runs, reports progress to the tree, and shuts down when done.

TaskTree Cloud

Run in secure managed VMs

Paid · per session

Secure, isolated virtual machines spun up on demand. No setup, no cleanup. Your agents work in a sandboxed environment with full developer tooling and ship results back to the tree.

  • Isolated, ephemeral VMs — one per session
  • Full developer tooling preinstalled
  • Network sandboxing & audit logs
  • Auto-shutdown on task completion
# from anywhere — phone, laptop, another agent
tasktree spawn --task "billing/invoices" --on cloud
→ provisioning vm-4f2 ...
→ session live · streaming to tree
SAS-client

Run on your own iron

Free · self-hosted

Install our open client on your own machine or server. Same tree, same orchestration, your hardware. Your iron, your rules.

  • Local machine, home server, or any VPS
  • Open-source client, no telemetry
  • Use existing dev environment & secrets
  • Connects back to the same TaskTree workspace
# once, on the box that will run sessions
curl -fsSL get.tasktree.dev | sh
sas-client register --workspace acme
→ ready · awaiting spawns
Three layers · one view

Product, Code, Infrastructure — linked by the same nodes.

The only PM system that gives cross-cutting visualization across all three layers, connected by the same task graph. No more switching between Linear, GitHub, and Terraform tabs to understand a single change.

Stop wasting time managing agents. Focus on the business — TaskTree handles the seams.

Product
· Business intent
layer.0
checkoutcartsave cardflowreceiptemail
Code
· Implementation
layer.1
initmaina1fmainb22feat/checkoutc4dd9efix/save-carde02f81PR #482 mergedcheckout-v2 → main12 files · +284 −5838 tests passing2 reviews · 1 comment
Infrastructure
· Runtime reality
layer.2
EC2api · t3.mediumus-east-1RDSorders-produs-east-1ALBcheckout-lbus-east-1S3assets-cdnus-east-1Lambdawebhook-fnus-east-1SQSinvoice-queueus-east-1CWlogsus-east-1IAMrolesus-east-1deploy v2.41.0 · canary 10% → 100%10%
Why it's different

Most "AI PMs" bolt a chatbot onto Asana.

TaskTree goes the other way.

TaskTree
Traditional PM
AI dev assistants
Built for AI as the primary operator
Persistent memory across sessions
Workflow phases as native primitives
Connects product → code → infra
Autonomous multi-task orchestration
Run sessions in secure cloud VMs
Run sessions on your own hardware
Who it's for

Built for the people who ship.

// team

Engineering teams

Running coding agents that should ship features end-to-end, not just patch the file in front of them.

// indie

Solo builders

Orchestrating autonomous AI workflows who've hit the wall on context, drift, and duplicated work.

// leadership

Founders & product leads

Who want to see business intent, code reality, and deployed state in one place.

// ops

Long-horizon work

Research, migrations, refactors, content production — anything that takes more than one conversation to finish.

Pricing

Priced by Compute Credits.

Deployment style decides where your agents run — our cloud or your own hardware. Compute Credits power what they do: the AI processing behind every task. That's the core product, so it's metered — not free.

Metered in Compute Credits

Every task TaskTree prepares, executes, tests, and finalizes draws Compute Credits (CC). You pay for work actually done — never for idle seats.

One blend of frontier models

No model tiers, no model picking. Every credit taps the same ensemble of frontier models under the hood, so every task gets our best.

Independent of deployment

Cloud VMs or your own hardware only decides where sessions run. The AI processing draws CC either way — hosting your own iron is free, the thinking is not.

Subscribe & save

Monthly Compute Credit packages

Start metered with no commitment, or subscribe to a monthly package and lock in a lower rate per credit. The bigger the package, the cheaper each credit.

Starter
10k CC · per month
Discounted vs metered

For steady solo work — keeping a single project tree moving.


  • Full frontier-model blend
  • Spend on cloud or your own iron
Join waitlist
Standard
20k CC · per month
Better rate per credit

For an active project with agents running across the tree most days.


  • Full frontier-model blend
  • Spend on cloud or your own iron
Join waitlist
Scale
50k CC · per month
Best rate per credit

For teams orchestrating many agents and long autonomous runs.


  • Full frontier-model blend
  • Spend on cloud or your own iron
Join waitlist

TaskTree is not released yet — exact credit rates and package prices land at launch. Join the waitlist to get pricing first.

Limited early access · Q3 2026

Your AI agents are ready to ship.
Get them a place to work.

Join the waitlist for early access. We're rolling people in cohort by cohort.

We'll never share your email.