This video walks through adding your first model, and verifying everything works by chatting with the agent.

Orpius Quick Start

Get from “installed” to “first automation” in about 10 to 20 minutes.

Who this is for

This Quick Start is for anyone new to Orpius who wants a fast, practical introduction using the Orpius Console.

If you need deeper detail (security model, architecture, custom tools, SDK integration), use the In-depth Technical Guide.

What you’ll achieve

By the end, you will have:

  • Installed the Orpius Console and connected to your Orpius Server
  • Added at least one language model provider
  • Run a first chat with Orpius
  • Created one automation (a Schedule or an Event)
  • Stored one Secret and referenced it safely

Key ideas (simple mental model)

You only need a few concepts to get started:

  • Organisation: your company or team boundary.

  • Space: a workspace for one project. Most things live inside a space.

  • Agent: an AI “user” that can chat and perform work.

  • Tools: capabilities an agent can use (scheduler, notifications, web retrieval, code execution, etc).

  • Schedule: runs an activity at a time/interval.

  • Event: runs an activity when triggered (often by an HTTP call).

  • Operation: A secure entry point to an agent that allows agent-to-agent communication and delegation of work.

    It can be used:

    • By other agents to delegate work (agent-to-agent communication).
    • By your application or external systems when integrating Orpius into your software or services.

  • Secrets: store API keys and tokens safely, so they are not exposed to model providers.

That’s it. Everything else is a refinement of these.

In simple terms: agents do the work, tools give them abilities, schedules, events, or operations decide when that work runs, operations can also connect agents to other agents, spaces organise everything, and secrets keep credentials safe.

1) Install and connect to Orpius

  1. Open the download link provided to you (it resembles https://<YourAppName>.app.orpius.com/console/).
  2. Download and run the installer.
  3. Launch the Orpius Console.
  4. Enter your Server URL.
  5. If you are a system owner, enter the Access Key.

If you are using the Orpius trial environment, no additional setup is required. Simply create your account and sign in when prompted.

If this is your first launch as the system owner, you’ll be prompted to complete initial setup (including email server settings and creating your account).

2) Create your account and open a Space

  1. Create your account when prompted (username, email, given name, surname, password).
  2. After sign-in, you’ll see your available spaces in the Space Selector.
  3. Open the default space, or create a new one for this Quick Start (for example, “Getting Started”).

Tip: Use a dedicated space for experiments so you can delete it later without affecting anything else.

3) Add a large language model (LLM)

Orpius connects to the model provider you choose. You can configure multiple models and assign different models to different agents.

  1. In the Console, open Models from the sidebar.
  2. Select Add (+).
  3. Enter the provider details (for example, OpenAI, Azure OpenAI, Google Gemini).
  4. Save.

If you are not sure which model to start with, pick one you already have an API key for. The goal here is simply to confirm inferencing is working.

In our testing, Gemini 3 Flash has performed reliably across a wide range of tasks while remaining cost-efficient.

4) Assigning models to built-in agents

Orpius includes two built-in agents:

  • Orpius – The interactive agent that receives your instructions through chat and coordinates tasks.

  • Phaedra – The agent that performs the tasks in the background.

By default, both agents use the first model you configure. If you add additional models, you may want to assign specific models to each agent to control behaviour, performance, or cost.

To change the model used by an agent

  1. Open Agents from the sidebar.

  2. Select Orpius or Phaedra.

  3. Choose the model you want the agent to use.

  4. Save your changes.

Tip: You can change agent models at any time as your workflows evolve.

5) Start your first chat

  1. Open Chat.
  2. Send a simple prompt, for example:

“Hello. Please tell me what you can do in this space.”

You should receive a response immediately. If you do, you have confirmed: Console connectivity, authentication, and model configuration are all working.

If you get an error, check:

  • your model configuration
  • your provider key
  • network access from the Orpius Server to the provider endpoint

6) Create your first Secret (safe credentials)

Secrets let you reference API keys in prompts without sending the actual values to the model provider. You will use references like:

<%=Key:WeatherKey%>

  1. Open Secrets from the sidebar.

  2. Select Add (+).

  3. Enter:

    • Token: WeatherKey (example)
    • Secret Value: your API key
    • Description: what it is used for

  4. Save.

Now you can safely reference it in prompts and activities.

This is the quickest way to see Orpius do unattended work.

A simple scheduled task: write a file and notify you

  1. Start a new Chat.
  2. Paste a clear instruction like this:

"Every day at 09:00, create or update a file named DailyNote.txt by appending today's date and one short sentence. After writing to the file send me a notification."

  1. Orpius will create the schedule.
  2. Open Schedules to review it.

Notes:

  • If you don’t specify repetition or end conditions, schedules may run indefinitely.
  • If notifications aren’t delivered, check whether the email server is configured and whether notifications are enabled.

8) First automation option B: Create an Event and trigger it

Events are how external systems commonly trigger Orpius. They can also be used from one task to trigger another task.

Create an event via chat

  1. Start a new Chat.
  2. Ask Orpius to create an event:

“Create an event named QuickStartEvent that appends a line to EventLog.txt with the current date/time and the words ‘Event triggered’. Then notify me.”

  1. Open Events and find QuickStartEvent.
  2. Expand the event URLs to copy URL for Access Key 1.

Trigger the event

Trigger it via HTTP GET or POST from a browser, curl, Postman, or your application.

  • If you paste the full event URL into a browser, it will usually trigger via GET.
  • Or use curl (example format):
curl "<PASTE_EVENT_URL_HERE>"

After triggering:

  • Check Files for EventLog.txt
  • Check Notifications for the message

Tip: Events can also accept query string parameters. Keys prefixed with _ are tool-only parameters; others may be passed to the agent as well, depending on how your event is set up.

9) A safe and cost effective delegation pattern using Operations

If you are integrating Orpius into your application, Operations are the standard entry point.

However, they are also a powerful way for agents to delegate specialised work to other agents. They help you reduce cost, because specialised sub-tasks can be handled by a more suitable or lower-cost model instead of using a single high-capability model for everything. This Quick Start focuses on this delegation pattern.

If you are integrating Orpius directly into your software or services, see Using Operations to Connect Your Application to AI Agents in the GitHub Technical User Guide.

Example: Using Operations for Receipt Processing

For example, imagine you need to process a large number of image documents, analyse them, and extract structured data from each file. The following example shows how you could use an Operation to delegate specialised work to another agent.

You might design this as follows:

Step 1: Create an Operation (This Operation is responsible only for extracting data from receipt images.)

  1. Open Operations.

  2. Select Add (+).

  3. Enter Operation name for exampe: FileContentEnquiries

  4. Choose the Agent that will handle requests. This can use a cost-efficient model (for example, Gemini 2.5 Flash), but you should always test whether the model is capable of handling the task. You can switch to a more capable model if needed.

  5. Enter Instructions to Agent:

    "You will receive a request to extract data from a financial receipt represented as an image' Use the ImageAnalysis tool to understand the content of the receipt and extract the data. Return the extracted data as JSON".

  6. Enable the Image Analysis tool (Note: only enable the Tools the operation genuinely needs, tools are disabled by default for safety).

  7. Save, then copy the External ID.

Step 2: Use Chat to create the Scheduled Task

  1. Enter instructions to Agent:

"Every day at 9:00 AM:

  1. List all files in the Receipts folder.
  2. If the folder is empty, send a notification to me.
  3. For each file:
    • Call the internal Operation (paste the External ID) using the file path as input.
    • Extract the data from the JSON response.
    • Append the extracted data (Date, Item, Price, Tax, Total, Category, and filename) to Ledger.xlsx.
    • Move the processed file to the Processed folder.
  4. Calculate the total sum of the Total column for the processed batch.
  5. Send a notification to me with the total sum".

What happens behind the scenes

  • The built-in agent manages the workflow and file handling.

  • A custom agent performs receipt analysis via the Operation.

  • Each agent can use the most suitable model for its task.

Common pitfalls (and how to avoid them)

  • “My schedule is running too often.” Always specify the interval clearly and set repetition limits where appropriate.

  • “I referenced a secret but it didn’t work.” Confirm the Secret exists and that you typed the token name correctly in <%=Key:TokenName%>.

  • “No notifications arrive.” Confirm the email server is configured (system owners) and that the Notifier tool is enabled where needed.

  • “My event keeps re-creating itself.” Avoid phrasing event activity instructions as “create an event” or “schedule an event”. Use immediate action wording.

Next steps

Once you have the Quick Start working, the usual progression is:

  1. Create a custom agent for one specific job (clear persona + instructions).

  2. Enable only the tools required for that job.

  3. Add one integration surface:

    • an Event endpoint for webhooks, or
    • an Operation for application-to-agent requests.

  4. Add auditing and retention policies appropriate to your environment.

When you’re ready, move on to:

  • In-depth Technical Guide (architecture, security, retention, auditing)
  • Development Guides (custom tools, SDK, operations integration)