Decisions as a Service

The Decision Runtime is Helixor's Embeddable AI Decision Engine. Turn a written policy into a versioned decision pack. Every call returns an action from the pack's closed set, the rules that passed or failed, a remedied payload and a receipt hash you can recompute. Evaluation calls no model and generates no language.

AvailableDecision Runtime 0.3.1

Try it now

Run a decision in the playground and read the action, the rules behind it and the receipt, or play search and games against the engine. No install and no license.

Three ways to call it#

Where it runs	What you do	Read
In your process	Load a pack into the Decision Runtime and call `evaluate()`. No network call, offline, with a license verified locally.	Evaluating decisions
As your own service	Serve a pack over REST, Server-Sent Events or WebSocket so services in other languages can call it, or run it as a sidecar.	Serving decisions
Hosted	Send a decision that needs more than local rules to the hosted Helixor reasoning API. It answers, abstains or refuses, and returns `p_correct`, its estimate of whether the answer is right.	Escalating to hosted reasoning

Start here#

Getting started

Quickstart

Try hosted samples, explore the demo repository locally, then build your own pack.

Getting started

Core concepts

Packs, playbooks, rules, actions, remedies, receipts and licenses.

Task guides

Common decision tasks

Route requests, reject what you do not handle, answer only when reliable, and serve decisions as an API.

Domain packs

Start with a business problem

Support routing, shipment release and purchase review, from policy to tested decisions.

Tutorials

Step-by-step builds

Your own pack, outcome memory, the reasoning service and the gated learning loop.

Reference

APIs and formats

SDKs, playbook and pack formats, the HTTP API, the CLI, errors and configuration.

Architecture

Run it in production

Deployment patterns, the six pillars and the production checklist.

Benchmarks

Measured results

Latency, throughput and decision quality, with the scripts to reproduce them.

Building an agent that makes these decisions in a conversation or a multi-step task? See the Agent Framework. Need a route or a roster rather than a single decision? See Solver as a Service.