Airia Agent

Airia Agent is a tenant-wide system agent managed and versioned by Airia. When enabled, it appears in the Chat interface for every user you grant access to, replacing the need for users to discover and select individual agents themselves. Unlike agents you build in Agent Studio, Airia Agent is automatically provisioned and maintained by the platform. You control what it can do — which tools it can call, which agents it can delegate to, who can access it, and how it presents itself — but Airia handles the underlying agent logic and releases updates over time.

​

How it works

When you enable Airia Agent, the platform provisions a hidden project and deployment in the background. From that point on:

• Users with access see Airia Agent in their Chat sidebar instead of a list of catalog agents
• You connect tools, MCP servers, and sub-agents to extend what it can do
• You optionally customize its behavior with a system prompt
• Airia publishes periodic updates with new capabilities; you choose when to apply them

Users without access are routed to a fallback catalog agent of your choice.

​

Token consumption & cost

Every Airia Agent conversation consumes tokens beyond the visible chat text. Some of that consumption comes from Airia’s own agent logic, and some comes from choices you make when configuring the agent:

• Airia’s base system prompt — Airia Agent ships with a platform-managed system prompt that is not visible to you or your users. It is sent to the model on every turn alongside any custom instructions you add, so it contributes to the token footprint of every message regardless of your configuration.
• Your system prompt — custom instructions are appended to the base prompt and sent on every turn, so longer prompts increase the token footprint of each message
• Tools & MCP servers — each connected tool or MCP server adds its definition to the model’s context, and tool calls consume additional tokens for both the call and its response
• Tool response history — the Save tool responses to history setting, when enabled, appends full tool call inputs and outputs to the conversation history sent to the model on every subsequent turn. This can meaningfully increase token consumption on longer conversations, especially with tools that return large payloads.
• Sub-agents — delegating to a sub-agent triggers its own model call with its own system prompt, tools, and context, on top of Airia Agent’s own consumption

Because of this, token consumption and cost can vary drastically from one conversation to the next depending on factors like the user’s prompt, how long the conversation runs, and how many tools or sub-agents get invoked along the way — and some of that variance comes from parts of the agent you don’t control. Use Feeds to monitor actual consumption and Billing to track cost.

​

Settings overview

All configuration lives at Settings → Airia Agent. The page is organized into tabs:

​

Required permissions

​

Recommended setup sequence

New to Airia Agent? Follow this order to avoid revisiting steps:

1. Enable Airia Agent — provision the agent for your tenant
2. Access & Privacy — decide who can use it and set a fallback agent
3. Extending Airia Agent — read this before configuring tools or sub-agents
4. Tools & MCP and/or Sub-agents — wire up capabilities
5. System Prompts — customize behavior for your organization
6. Compute Sandbox and Skills Repositories — optional, for advanced tool execution isolation

---