AI Assistant Settings (Ora)

:::info Screenshot needed — light/dark 📸 settings-window-light.png and settings-window-dark.png — Settings Window, custom title bar :::

The Ora tab (header text: Ora) configures the built-in AI Assistant. Changes are saved when the Settings window closes — the AI session resets automatically, so you do not need to restart the application.

Profile Configuration



Ora is driven by one or more profiles. Each profile pairs a provider with its connection details and the models it can use. Profiles come in two kinds:

• Built-in provider profile — backed by a provider managed through your subscription or license. These show "Managed by your subscription" and cannot be edited or deleted from this tab; their models can be restored to defaults.
• Custom Endpoint profile — "Full control over URL, protocol, and models." You supply the connection details yourself.

Action	Icon	Description
Add Profile		Opens the Add Profile wizard (Choose Provider → Configure Endpoint → Profile Details).
Edit profile		Opens the Edit Profile dialog for a custom profile. Not available for managed profiles.
Delete profile		Removes a custom profile after confirmation. Not available for managed profiles.

If no profiles exist, the tab shows: "No profiles configured. Click 'Add Profile' to get started."

note

The active profile is selected from the Ora assistant panel's model picker, not from this tab. Adding a profile here does not automatically make it active.

Provider & Connection

These values are set when you add or edit a profile (in the wizard or dialog):

Field	Description
Provider	For a built-in profile, the provider name (e.g. Aethrix-managed). For a Custom Endpoint, this is "Custom Endpoint" with full control over URL, protocol, and models.
API Key	Your personal API key for the provider. Masked with password characters. For managed profiles it is handled by your subscription.
Base URL	Custom endpoint URL for private or on-premise LLM deployments (Custom Endpoint profiles). Built-in providers need no URL from you: Qwen, Kimi and GLM operate a separate endpoint per region and their keys do not cross regions, so Studio dials the address that matches your account's residency — the international endpoint for Global / EU / US, the China endpoint for China. The provider card shows the address the next request will actually use.
API Protocol	Wire protocol for the endpoint — set it to match the provider you have a key for. OpenAI-compatible is the Chat Completions dialect most providers speak; Anthropic Messages is Claude's and DeepSeek's; OpenAI Responses is OpenAI's newer API. Built-in providers arrive pre-set.
Models	The models available to the profile. Each model has a Display Name, a Model ID, and an optional Thinking effort, which defaults to High. Which levels you can pick depends on the model: some accept Off, and a model that would think whatever a request says is offered without an Off item rather than with one that does nothing. A model you add declares its own levels: tick the ones your endpoint reads, and the effort menu shows exactly those — leave one unticked and Studio never sends it. Each row also states its API Type — Chat Completions, Anthropic Messages, or Responses — pre-filled from the provider you chose and changeable per model, because a model's dialect is a property of the model. A model from the built-in gallery arrives with the levels that model is known to accept; you can adjust them for this profile, and Restore default models brings the gallery's list back. Use Add Model to append, Remove to delete, and Restore default models to reset a built-in provider's model list to what the provider currently declares.

• What it does: Defines which LLM Ora talks to and how it connects.
• When to use: Add a Custom Endpoint profile to use a self-hosted or on-premise model, or to point at a provider not built in. Set the API Protocol to match the endpoint's wire format so requests are encoded correctly.
• How to set: Click Add Profile, choose a provider (or Custom Endpoint), configure the endpoint (Base URL, Protocol) and profile details (name, API key, models), then finish. Edit an existing custom profile with the edit (pencil) button to change these later.

Verified models

Ora works with any model that speaks one of the three published dialects — Chat Completions, Anthropic Messages and OpenAI Responses — so most models behind one of them work. These are the ones we have tested in Studio:

Provider	Models
OpenAI	GPT-6 Luna, GPT-6.1 Sol
Anthropic	Claude Sonnet 5.5, Claude Haiku 5.5
DeepSeek	DeepSeek Flash
Qwen	Qwen3.8 Max, Qwen3.8 Flash

To use any other model, add a Custom Endpoint profile, pick the protocol your provider speaks and type the model id. Studio sends the plain request shape of that dialect, so what you get is what your provider implements.

Behavior



Require Tool Confirmation

• What it does: When enabled, the AI shows a confirmation card before executing any action in Aethrix Studio (e.g. modifying requests, running tests). When disabled, actions execute automatically.
• When to use: Keep it enabled in shared or production-adjacent environments so you approve each action. Disable it for trusted, hands-off automation workflows.
• How to set: Open the Ora tab → Behavior → Require Tool Confirmation, toggle the switch.

Max Conversation History

• What it does: Number of conversation turns kept in memory. Lower values reduce token usage. Set to 0 for unlimited.
• When to use: Lower it on long sessions or token-constrained plans to bound memory and cost; raise or set to 0 when you need the assistant to keep full context.
• How to set: Open the Ora tab → Behavior → Max Conversation History, type the number of turns, changes auto-save on close.

Font Size

• What it does: Base font size for the Ora panel's text — the reply, the streaming text and the thinking text. Options: 12, 13 (default), 14. Captions and chips keep their own smaller sizes, and headings in the reply step up from this size. Settings → Appearance → Interface Scale scales the panel on top of this choice.
• When to use: Increase it for readability on high-DPI displays, or decrease it to fit more content.
• How to set: Open the Ora tab → Behavior → Font Size, pick a size from the dropdown — the panel updates immediately.

Specialist Iteration Budget

• What it does: Maximum tool-call iterations per specialist dispatch before the task is cut and split (0 = unlimited).
• When to use: Lower it to bound how much work a specialist (e.g. TestBuilder) does in a single dispatch; when the budget is reached Ora pauses and asks whether to continue or stop.
• How to set: Open the Ora tab → Behavior → Specialist Iteration Budget, set a value between 1 and 100, changes auto-save on close.

Orchestrator Iteration Budget

• What it does: Maximum tool-call iterations per user turn before the loop is aborted (0 = unlimited).
• When to use: Lower it to keep a single user turn from running too long; when reached, Ora warns you that the turn was cut rather than surfacing a system error.
• How to set: Open the Ora tab → Behavior → Orchestrator Iteration Budget, set a value between 1 and 150, changes auto-save on close.

Interaction Style



Pick a style by who you are, not by how chatty you want Ora to be. All three use the same product vocabulary — ${Variable} references, field names, mapping rules, data-driven cases, assertion types appear in every style — and none of them shows Ora's tool names or internal parameters. What a style changes is what gets explained, and how much Ora does without asking.

• What it does: Three mutually exclusive styles:
  • Guided (default on first launch) — for anyone new to the interface or to test design. Ora translates both, defines a term the first time it appears, builds the test end to end, and finishes by telling you what to look for on screen. Before a send it is not sure of — an address it had to supply, a value it filled in itself — it asks which you want: send as it stands, or hold while you look the request over.
  • DEV — for readers fluent with APIs and payloads. Ora keeps the exact field names, variable references and status codes, skips preamble, and acts without asking — stopping only before destructive changes. Test-design techniques still get a one-clause explanation of what they cover.
  • QA — for the person who owns the business rules and the coverage. Ora leads with behavior and risk, proposes the technique with its coverage rationale before generating anything, keeps interface-level detail to what changes a test outcome, and puts the same send-or-review choice to you rather than running a request whose consequence it cannot justify.
• When to use: Guided while learning the product; DEV for fast expert workflows where you want the literal names to act on; QA when the question is coverage and business behavior rather than payloads.
• How to set: Open the Ora tab → Interaction Style, select one of the three cards; the choice is saved with your settings and the current conversation resets. You can also switch mid-conversation with /personality (e.g. /personality DEV), which shows the three names and the one in force.

MCP Servers



Model Context Protocol (MCP) servers expose additional tools to the AI agent.

Adding an MCP Server

Click Add Server and configure:

Field	Description
Name	Display name for the server
Transport	stdio (local process) or sse (Server-Sent Events over HTTP)
Endpoint / Command	For stdio: executable path. For SSE: HTTP URL.
Arguments	Space-separated command-line arguments (stdio transport)
Enabled	Toggle to activate/deactivate the server without removing its configuration

You can also click Edit JSON to view and edit the full server list as JSON, then Done to apply it.

Managing Servers

• Click Remove on a server to delete it.

• Changes are saved when you click the main Save button; the AI session resets automatically to pick up MCP server changes.

• What it does: Registers external MCP tool providers so Ora can call their tools.

• When to use: Add an MCP server when you want to extend Ora with custom or third-party tooling (e.g. internal APIs, databases, or CLIs).

• How to set: Open the Ora tab → MCP Servers → Add Server, fill in the fields, ensure Enabled is on, changes auto-save on close.

See Also

• Getting Started with AI Assistant — using Ora in your workflow
• General Settings