ThinkWork

ThinkWork is the open Agent Harness for Business — Reliability, Efficiency, Security, and Traceability built into the runtime, not bolted on. Self-host on your AWS, run it with us, or wrap either with services. Agents do the work, Spaces frame the work, Threads carry the record, Memory carries context, integrations bring in the world, automations time the work, and controls keep it accountable — all inside the AWS account your team already runs.

Getting Started View on GitHub

What ThinkWork is

ThinkWork is the open Agent Harness for Business — a Terraform module that stands up a production-grade agent runtime inside your VPC, keyed to your Cognito, running on your Bedrock endpoint, with agent state in your Aurora cluster. The harness — Agents, Spaces, Threads, Memory, Integrations, Automations, Control — is what’s opinionated; the infrastructure is yours. Reliability, Efficiency, Security, and Traceability are built into the runtime, not bolted on as an afterthought.

If ThinkWork the company disappears tomorrow, your deployment keeps working.

One harness, three deployment models

ThinkWork (open)

Self-host the open Agent Harness on your AWS. Apache 2.0, community-supported. The full product, no operating partner. Five commands and you own a production-grade agent runtime. The rest of this site is the manual for it.

ThinkWork for Business

Same harness, operated by us — deployed into your AWS account, with managed updates, priority support, and the operating discipline so your team focuses on the workflows, not the runtime. See plans →

ThinkWork Enterprise

Strategy, pilot launch, managed operations, and workflow expansion services on top of either ThinkWork (open) or ThinkWork for Business. Sales-led; scoped up front. See services →

These docs cover the full product across all three tiers. The runtime is identical between them — only who operates it differs.

Where to go from here

These docs are organized into four branches, each answering a different question:

Architecture

Start here. The harness mechanics — PPAF agent loop, the four operating guarantees, deployment topology, and how the core components fit together. Read more →

Components

The core components of the harness — Agents, Spaces, Threads, Memory, Integrations, Automations, and Control. Each page follows the same skeleton: why, what, how to configure, common patterns. Read more →

Configure

Runbooks for standing the harness up and tuning it — deploy paths (greenfield AWS, BYO infrastructure, configuration reference) and authoring guides for skill packs, integrations, evaluations, and compounding memory operations. Read more →

Reference

The technical reference surface — GraphQL API, the React Native SDK, and the operator-facing applications (Admin, Mobile, CLI). Read more →

Core components of the harness

Every concept page in these docs follows the same skeleton: why the component exists in the harness, what it does, how to configure it, common patterns, and cross-links into the rest of the system.

Agents

The decision-making layer. One tenant platform agent runs through Space-scoped context, with folder specialists and connected runtimes when you bring your own agent service. Read more →

Spaces

Contextual workrooms. Spaces hold local workspace files, tools, memory, triggers, channels, and runtime policy for the tenant platform agent. Read more →

Threads

The harness’s perception and history layer. Every conversation, task, automation, and integration event lands in a thread with history and audit trail. Read more →

Memory

The harness’s context layer: memory, compiled pages, knowledge bases, workspace files, approved tools, and source routing above any one backend. Read more →

Integrations

The harness’s I/O surface. External channels, events, and tools — Slack, GitHub, Google Workspace integrations plus MCP tools for remote capabilities. Read more →

Automations

The harness’s scheduled-trigger surface. Scheduled jobs, event-driven triggers, and routines on your schedule. Step Functions under the hood. Read more →

Control

The harness’s governance layer — the four operating guarantees enforced in code. Guardrails, budgets, evaluations, and audit. Read more →

Quick start

The fastest path is self-hosting the open Agent Harness in your own AWS:

npm install -g thinkwork-cli
thinkwork login
thinkwork init -s dev
thinkwork deploy -s dev
thinkwork doctor -s dev

Five commands, one AWS account, and you own a production-grade Agent Harness — open and yours, not rented from a black box. Full walkthrough in Getting Started; the harness mechanics — PPAF agent loop, operating guarantees, deployment topology — live in Architecture.

If you’d rather not run the runtime yourself, ThinkWork for Business is the same harness operated by us, and ThinkWork Enterprise wraps either with services.

If you’re new to ThinkWork, read the Concepts section in this order: Agents, Spaces, Threads, Memory, then Integrations.