Open infrastructure for AI work
AI infrastructure in your AWS.
Without the black box.
A production-grade harness for AI work, inside your own AWS account. Portable memory. Durable threads. Control that stays under your control.
Five-command deploy
From clone to a live harness in your account.
Runs in your AWS
No SaaS control plane. Your AWS account, your data.
Open source core
Inspect it, fork it, ship it.
Threads · Memory · Control
Built in, not bolted on.
Why ThinkWork
Own the harness, not just the prompt.
The easy path to serious agent infrastructure shouldn't cost you ownership of the system doing the work.
Easy infrastructure
Agent infrastructure is hard. ThinkWork makes it a five-command path — without forcing you into a hosted black box.
Runs in your AWS
The runtime lives in your AWS account. The work record stays in your systems. Governance and audit stay with your infrastructure, not a third party.
Portable memory contract
Memory is a harness-owned context layer, not a backend vendor's API. Inspect it, export it, keep it.
System model
Threads run the work. Memory carries it forward.
Four primitives, one control surface wrapping them. Simple enough to hold in your head, strict enough to ship against.
Threads
The system of record for AI work. Every request, action, and outcome lives here.
Memory
A portable, harness-owned context layer that carries work forward between threads and agents.
Agents
Managed or self-hosted. They operate inside the same thread and control model.
Connectors
Approved system access — without the connector becoming the product contract.
Control wraps the whole system
The wedge
Your memory. Your runtime. Your AWS.
ThinkWork's two hardest-hitting differences aren't features — they're where the system lives and who owns the context.
Portable memory
- Memory is a harness-owned context layer, not a backend vendor's API.
- Portable and inspectable — you can read it, export it, and move it.
- A stable memory contract above pluggable engines. Hindsight and AgentCore are adapters, not the product.
Your AWS account
- ThinkWork deploys into your AWS account. No SaaS control plane.
- The work record stays in your systems. The runtime runs in your AWS account.
- Governance, audit, and network controls stay where your team already manages them.
End-user app
Your users get a real mobile app.
The operator story lives in the admin web. The user story lives in a native-feeling iOS app built on the same threads, agents, and connectors the rest of the system uses.
Threads
Tasks
Assigned work, one place
Chats, automations, emails, and external tasks flow into a single inbox. Nothing asking the user to keep five tools open in parallel.
Native GenUI, not markdown
Task cards render as native mobile components — fields, actions, activity — from a bounded block grammar the server controls.
Realtime by default
Webhook events, agent turns, and status changes land on-device within seconds over AppSync. No pull-to-refresh dance.
On TestFlight today
iOS first, live for early users. The Expo codebase is cross-platform; Android ships when the iOS shape is stable.
Quick start
Five commands. One AWS account.
Real infrastructure, not a hand-wave. Clone, configure, deploy — in your account. Full setup steps and backend options live in the docs.
Full getting started01$ npm install -g thinkwork-cli02$ thinkwork login03$ thinkwork init -s dev04$ thinkwork deploy -s dev05$ thinkwork doctor -s devThe harness stays yours.
Deploy a production-grade agent system inside your own AWS account. Keep the runtime. Keep the memory. Keep the work record.