The runtime for agent-native products

Build the product around the agent.

Orage is a self-contained runtime for turning expert methods into AI agent products people can use. Experts shape the experience; Orage keeps behavior, tools, data, channels, and interfaces inside one product boundary.

See inside an instance Read the technology

Unit: One instance, one complete product
Contains: Agents, skills, tools, memory, files, identity, channels, and site
Runs: As one deployable box
Controlled by: The creator, not the model provider

Anatomy of an Orage instance

One boundary holds the whole product.

The model can change. The product still needs behavior, authority, state, distribution, and a record of what happened.

Orage instance Isolated product boundary

Creator control

Method
Skills
Tools

Agent runtime

Agent
Memory
Files
Identity

Product surfaces

Public site
Web app
Channels
API

Durable history

Agent runs, tool calls, and delivery events produce durable records that can be inspected and used to rebuild state.

The package is designed to keep the same instance boundary across managed and creator-controlled deployment. Supported migration is still product work.

Dated proof

Parts of the loop already run.

16 Aug Creator edited a skill in the browser. The next agent reply followed it.
18 Aug A controlled WhatsApp path received, ran, replied, delivered, and read.
30 Aug A hosted agent used four creator-approved specialist tools in one live turn.

Read the scope of each test

From expertise to product

The creator shapes it. People use it.

Orage is the technical partner between a specialist's judgment and the experience their audience receives.

Creator workspace

The coach defines the method.

skills/rider-check-in/SKILL.md Saved

Before suggesting a session, ask how the rider feels and what they want from today.

Voice: Calm, direct, and specific. Never pretend to replace the coach.

Expert shapes the method Orage runs the product Audience gets the experience

Audience experience

The rider gets useful judgment.

My legs feel heavy. Should I still do intervals?

Coach

Before I change today's session, I need two checks. How did you sleep, and was either of your last two rides hard?

Five hours. Yesterday was hard.

What returns to the creator

The product can improve without losing its author.

Suggested change: ask about sleep before changing workout intensity.

Illustrative cycling product. It shows the intended creator-to-audience loop.

Hosted pilot

Astrology guide

A distinct voice with a creator-approved specialist tool set.

Requested product

Lean management mentor

An expensive specialist method made useful between human sessions.

Why a runtime

The model is one part. Orage holds the product together.

A prompt can describe an agent. A product also needs a safe place to run, remember, reach people, and remain under its creator's control.

Creator control: Behavior lives in editable skills and policies, not a hidden platform prompt.
Product distribution: The website, signed-in app, channels, and API belong to the same instance.
Operational memory: Files, personal memory, usage, and a durable event tape survive the next model call.
Deployment boundary: The runtime, head, and data paths ship as one box with explicit authority.

Technical shape

Small enough to understand. Complete enough to operate.

Runtime: Node 24 and Hono
Agent loop: Provider-neutral, streaming first
State: Durable tape, files, and encrypted stores
Authority: Identity, permissions, tool policy, and secret brokering
Heads: Reference web app and replaceable public site
Target: Local, Docker, cloud VM, or customer premises

The box is self-contained by design. Supported self-service migration between hosts is still product work.

Turn expertise into software people can use.

Shape the method. Own the product. Let Orage run what connects them.

Open Orage Read the technology