Hyperstruck Engine

The runtime where experience is native

A full reasoning system, plan, execute, reflect, revise, built around the experience layer rather than bolted onto it. It is the reference implementation of Hyperstruck, and the shortest path to an agent that learns your business from day one.

The fork

When to choose it

Add the layer to your stack

You already run LangGraph, OpenAI Agents SDK, CrewAI, Claude Code or Cursor, and it works for you. Plug Hyperstruck in underneath; nothing gets ripped out, and your agents start earning experience on the stack you have.

Build on the Engine

You are building fresh, you want one coherent system rather than assembled parts, or you need the whole thing delivered and governed for an enterprise. The Engine gives you the runtime with experience already inside.

The runtime

What is inside

Planning

Goals decompose into dependent steps with milestones, not a single long prompt.

Execution

Structured tool calls with retries, recovery and dependency handling.

Reflection and revision

Work is challenged with evidence and known lessons before it ships, and revised when it falls short.

Human-in-the-loop gates

Approval points at plan, step and tool level, for the actions that warrant a person.

Experience, native

Lessons are extracted at milestone boundaries during the run, not just at the end, and the rules the record holds are in force at every decision point. This is where experience goes deepest.

Delivery

Hosted or self-hosted

Run it hosted on Hyperstruck, or self-hosted in your own Kubernetes environment with the full stack bundled. Licensing and dedicated support for enterprise deployments.

The Engine competes with the frameworks the layer plugs into, and that is deliberate: it shows what the platform makes possible when experience is built in rather than bolted on. If you already run a framework you like, keep it; the layer exists for exactly that.

Build agents that improve with experience

Talk to us about whether the Engine or the layer fits your team.