# Try ORYH locally

No sales call is required to explore the open-source product. This is a
self-hosted developer preview, not an anonymous hosted sandbox or a free
model service. Use fictional data and your own agent/model access.

## 1. Start the business system

Requirements: Docker Compose, Git, enough disk space, and an agent that can
read Skills and make HTTP requests. Follow the maintained public installation:

- Source and Docker setup: https://github.com/AIE-enginehub/Oryh
- First boot: https://github.com/AIE-enginehub/Oryh/blob/main/docs/manual/first-boot.md

Clone the source into a fresh directory. To keep the public Compose gateway
on loopback, run from the repository root:

```sh
git clone https://github.com/AIE-enginehub/Oryh.git oryh-local-trial
cd oryh-local-trial
ORYH_CONSOLE_PORT=127.0.0.1:8080 docker compose up -d --build
```

First boot creates the workspace and its administrator. Keep the first-boot
credentials private; do not paste them into Chat, screenshots or public
support requests. Open http://127.0.0.1:8080/ and follow the first-boot guide.
Do not expose this evaluation instance or reuse demonstration credentials.

The public workspace starts empty. The videos at https://oryh.ai/demos use
a separately prepared fictional Northwind company; those prepared records,
internal validation tooling and hosted approval-flow skills are not part of
the public installation. Never point a data seeder at a production workspace.

## 2. Connect the agent you already use

Open http://127.0.0.1:8080/web/connect and install the credential-free bootstrap
Skill where your agent discovers Skills. Ask it to connect to this local
ORYH instance. Complete browser authorization using your own local account.
The agent receives its own scoped connection; do not paste keys into prompts.

The current connection options and client boundaries are documented at:
https://oryh.ai/docs#install

Codex, Claude Code, Hermes and OpenClaw are example compatible clients.
Evaluate your particular client and version. A cloud-hosted agent cannot
reach your machine's loopback URL; use a locally running client or a separately
secured deployment rather than publishing your demo database to the internet.

ORYH Client is optional. Its public source and setup are at:
https://github.com/AIE-enginehub/oryh-ai-client
If running it in Docker too, follow its documented server-network setup;
localhost inside a container is that container, not the ORYH container.

## 3. Try one draft before configuring a complete workflow

As the local workspace administrator, ask your agent to create fictional
master facts: Meridian Field Services Ltd; a calibration service listed at
USD 1,250; an on-site commissioning day at USD 850; and a Meridian rollout
project. Have the agent show the proposed writes and ask before saving.

Then, using a person with the appropriate capability, try:

> Prepare a draft quotation for Meridian: 8 calibration services at USD
> 1,187.50 each, plus 2 commissioning days at USD 850. Total USD 11,200.
> The 5% calibration discount is for annual renewal. Ask me for missing
> terms, show every proposed write, and wait for confirmation. Do not send it.

Or choose an unoccupied reporting week and ask for five 8-hour project entries:

> Prepare my timesheet for this completed reporting week, Monday to Friday,
> 8 hours per day on Meridian rollout: requirements, planning, proposal,
> customer review and handover. Check for an existing period first. Show all
> dates, tasks and 40 hours before saving anything. Do not submit yet.

Inspect the stored record after confirmation. Use your actual local person
and project names, and correct dates deliberately; never overwrite an existing
period to make an example look successful.

## 4. Evaluate approval routing separately

A complete workflow also needs separate submitter/approver identities, their
roles and capabilities, published company data rules and approval routes,
and a workflow-driving agent. The public product supports company-owned
definitions and Skills; it does not ship the cloud service's hosted flow
agent or the private validation bundle. Author/configure your own driver or
arrange a hosted pilot rather than expecting an automatic approval after
submission.

The approver's agent records the approver's decision. A workflow-driving
agent interprets the rules and advances the final document state. Those are
separate acts. Confirm saved actors, amounts, hours, approval facts and todos
instead of trusting the last sentence of a conversation.

See https://oryh.ai/demos for bounded local validation results and known
limitations, or https://oryh.ai/enterprise-pilot for a working session around
your own approval process.

Stop the local instance with `docker compose stop`. Do not use `down -v`
unless you deliberately intend to delete its local data permanently.