> ## Documentation Index
> Fetch the complete documentation index at: https://docs.praxis-ai.com/llms.txt
> Use this file to discover all available pages before exploring further.

# The Agent Workspace

> Where you run autonomous agents day to day — the agents page, the remote desktop, saved workflows, and the connections agents use.

Once a Digital Twin is eligible to run autonomous agents, the place you actually work is the **Agent Workspace**. Open it from the **Agents** button in the top navigation bar, or go to `/pria/agents`.

<Note>
  This page covers the workspace people use. The separate [Agents](/mdx/admin-guide/agents) page covers the **admin dashboard** — workspace status across your Digital Twin, heartbeat tasks, pending approvals and Shield events.
</Note>

<Info>
  **You won't see it until it's switched on.** The Agents button appears only when a platform administrator has made your Digital Twin eligible (**Autonomous Agent** in the twin's settings — see [Configuration](/mdx/admin-guide/configuration#autonomous-agent)) and you administer that twin. On a twin without it, the workspace sends you back to your conversation.
</Info>

***

## The agents page

The workspace is a sidebar beside a main panel. The sidebar lists what you can run and what you have run; the panel shows the conversation with whichever agent you pick.

### The sidebar, top to bottom

| Section             | What it holds                                                                                         |
| ------------------- | ----------------------------------------------------------------------------------------------------- |
| **Agents** (header) | The workspace name and a **pin**, which keeps the sidebar open instead of letting it collapse         |
| **Running**         | Agents working right now. Each row shows its name and a green dot; hover a row for **✕** to stop it   |
| **Available**       | The agents this Digital Twin offers. Each row has a name, a one-line description and a **Run** button |
| **History**         | Recently finished sessions, newest first, each with how long ago it ended                             |
| Footer              | **Remote Desktop**, **Workflows**, **Connections** — the three destinations below                     |

An agent in **Available** can show **Unavailable** instead of **Run**. That means the agent is switched on for the twin but something it needs is missing from the runtime; the row says what. Nothing is hidden from you just because a check hasn't run yet — if the platform has not yet observed the runtime, every enabled agent stays runnable.

<Tip>
  **Collapse the sidebar and you keep everything.** Unpinned, it becomes a narrow icon strip carrying your running sessions plus the same three destinations, in the same order. Hover it to peek at the full list without pinning it open.
</Tip>

### The main panel

* **Nothing selected** — "Select an agent to get started".
* **A running session** — the live conversation, where you can keep talking to the agent while it works.
* **A finished session** — the same conversation, read-only.

Picking a session from **Running** or **History** opens it in the panel. Starting one from **Available** deploys the agent and switches to it.

***

## Remote Desktop

Agents that need a real computer get one: a private cloud desktop you can watch and use in the browser. Open it from the sidebar footer.

The page opens on a status card. What you can do depends on what it says:

| Status              | What it means                                                           | What you can do                                          |
| ------------------- | ----------------------------------------------------------------------- | -------------------------------------------------------- |
| **Inactive**        | The autonomous Agent for your Digital Twin is not currently provisioned | Ask a platform administrator to enable it                |
| **Queued**          | Activation requested; the platform is preparing it                      | Wait — it moves on by itself                             |
| **Setting up**      | The workspace is being built                                            | Wait; the desktop appears when it's ready                |
| **Ready**           | The workspace is up                                                     | **Connect to Agent Desktop** opens the live screen       |
| **Sleeping**        | Idle, paused to save compute; your files are kept                       | **Resume** wakes it                                      |
| **Needs attention** | Setting up didn't finish                                                | The card shows the reason; ask an administrator to retry |
| **Stopped**         | The workspace was removed                                               | Ask an administrator to re-activate it                   |
| **Runtime missing** | Marked ready, but no desktop is actually attached                       | Needs reprovisioning — tell an administrator             |

<Note>
  The desktop connects only when you ask. Opening the page shows the status card; the screen itself is not streamed until you press **Connect to Agent Desktop**, and closing it drops the connection.
</Note>

***

## Workflows

A **workflow** is a saved agent job — an agent plus the instructions you keep giving it — so you can start it again without setting it up each time.

The page lists what you have saved, with **New workflow** at the top right. Each saved workflow can be run, edited, or opened to see what previous runs produced.

### Creating one

The editor has three sections, in this order:

<Steps>
  <Step title="Workflow">
    **Name** — what you'll call it in your list.
    **Agent** — which agent the workflow starts.
  </Step>

  <Step title="Instructions">
    **What should it do?** — the instructions saved with the workflow, used every time it runs.
  </Step>

  <Step title="Schedule">
    **How often it runs.**
  </Step>
</Steps>

**How often it runs** offers **Only when I start it** (the default), **On a repeating interval**
(with minutes between runs), **Once a day**, **Once a week** and **Once a month**. Daily, weekly and
monthly runs are set in UTC, so they do not shift with your clocks. A scheduled run starts **as you**,
using your saved connections, and behaves exactly like starting it yourself.

<Note>
  **Unattended runs are switched on per environment.** Your schedule is saved, and the next run time
  is worked out and stored with it — but a workflow only starts on its own where the platform has
  enabled unattended runs. Until then the Workflows page says so, and workflows run when you start
  them. Ask the Praxis AI team whether it is on for your deployment.

  Two things hold even once it is on: a run that is still going when the next one is due is
  **skipped** rather than stacked, and a scheduled run never gains rights you do not have.
</Note>

### Run history

Opening a workflow's run history shows each previous run and lets you read what it produced. A run started here behaves exactly like starting the agent yourself from the agents page — same agent, same connections, same billing.

***

## Connections

Everything an agent reaches outside Pria lives on one page, arranged as three tabs.

### AI Providers

The LLM subscription your deployed agents run on. Connect one and agents route their turns through it.

| Provider            | What connecting does                                                     |
| ------------------- | ------------------------------------------------------------------------ |
| **ChatGPT / Codex** | Lets agents make AI calls against your own ChatGPT or Codex subscription |
| **Grok**            | The same, against your xAI / Grok subscription                           |

Connecting finishes in a browser window, and the row then reads *connected*.

### Connectors

Connect your agent to external applications using connectors.

| Connector        | What it gives an agent                                                              |
| ---------------- | ----------------------------------------------------------------------------------- |
| **HubSpot**      | Read and propose changes to CRM records. Shows which portal is wired once connected |
| **Exa**          | Web search. Paste an Exa API key                                                    |
| **Notion**       | Reads your connected Notion workspace                                               |
| **Pria API key** | Lets an agent call Pria's own API as this Digital Twin                              |
| **Slack**        | Links your Slack account so you can talk to agents from Slack                       |

<Note>
  Some connectors are set up for the whole Digital Twin rather than for you personally — HubSpot, for instance, says *"Ask an instance administrator to connect HubSpot"* when you cannot do it yourself.
</Note>

**Linking Slack** issues a one-time code: press **Generate code**, then type `/synaps link <code>` in any Slack channel where the agent bot is present. Codes last five minutes and work once; **Refresh** issues a new one.

### Variables & Secrets

Live configuration your agents read while they work — an API key, an endpoint, an account identifier. Updates take effect on the agent's next connections call; nothing needs restarting.

Each entry has:

| Field              | Meaning                                                          |
| ------------------ | ---------------------------------------------------------------- |
| **Name**           | The name the agent reads it by. Capitals, digits and underscores |
| **Value**          | What it resolves to                                              |
| **Classification** | **Value** (readable) or **Secret** (write-only)                  |
| **Scopes**         | Where it applies — the desktop, the session, or both             |

<Warning>
  **A secret is write-only.** Agents can see that a secret exists, but its value is never returned to the workspace, and it is never shown back to you after saving. To change one, save a new value over it.
</Warning>

Entries can be disabled without deleting them — a disabled entry stops resolving immediately and keeps its history. Deleting one is permanent and asks you to confirm.

***

## Related

* [Agents (admin dashboard)](/mdx/admin-guide/agents) — workspace status, heartbeats, approvals, Shield events
* [Configuration](/mdx/admin-guide/configuration#autonomous-agent) — making a Digital Twin eligible
* [Single Model vs Agent Mode](/mdx/user-guide/interface/single-model-vs-agent-mode) — subagents, which are a different feature
* [Connectors](/mdx/admin-guide/connectors) — connectors for the twin's own conversations
* [Credits](/mdx/user-guide/credits) — how agent work is billed