AI Assistant (Ora)
Ora is the built-in AI assistant in Aethrix Studio. It helps you test APIs, build test automation projects, debug failures, and find documentation — all through natural conversation.
Ora runs as a multi-agent system: a single Orchestrator drives the conversation and dispatches focused sub-tasks to specialist agents. Capabilities and the conversation experience continue to evolve with each release.
Architecture
Ora uses a multi-agent system built around the Orchestrator and a set of specialists:
The Orchestrator owns the conversation, understands intent, and dispatches sub-tasks to specialist agents through the dispatch_to_specialist tool. Each specialist runs a single focused turn, streams its result back to you in real time, and returns a compressed summary to the Orchestrator, which composes the final response. Simple single-field request edits and documentation lookup are handled directly by the Orchestrator (they are not separate dispatched specialists); full request filling and test creation are dispatched to TestBuilder.
| Agent | Role |
|---|---|
| Orchestrator | Owns the conversation, parses intent, dispatches to specialists, and handles documentation lookup (search_documentation), simple request edits, context management, and view switching directly. |
| TestBuilder | Fills request bodies and creates test structure: suites, cases, steps, assertions, data sources, and data-driven setup. This is the default dispatch for any request-filling or test-creation task (it absorbed the former Request specialist). |
| Execution | Runs tests and sends requests, diagnoses failures, and proposes additional assertions. Also handles load/performance tests. |
| TestDesigner | Applies ISTQB techniques (equivalence partitioning, boundary value analysis, pairwise, decision tables, state transition, MC/DC, …) to design structured test cases. |
| Simulation | Configures service simulation responses with conditions: typed/plain responses, body filling, and conditional rules. |
Chat Interface
The Ora panel appears as a sidebar on the right of Aethrix Studio. It can also be detached into a separate window (see Getting Started).
Message Types
| Type | Visual |
|---|---|
| User message | Accent-colored bubble, right-aligned |
| Assistant message | Left-aligned with a role-colored accent bar; Markdown rendering |
| Tool notification | Centered chip showing the tool being called (e.g. "Calling: set_request_from_json") |
| Agent role badge | Compact label when the active specialist changes |
| Agent section | Collapsible group header for a specialist sub-task, tinted by the specialist's role color, with its token cost |
| Tool confirmation | Card with Approve / Deny buttons when confirmation is required |
| Continue / Pause | A confirmation-style card when an iteration budget is reached — Continue resumes with a fresh budget, Pause stops the loop (see Iteration Budgets) |
| Tool call group | Collapsible summary of the specialist's tool calls (✓ / ✗ per tool) |
| Context compaction | Notice shown when the Orchestrator summarizes older turns to free context |
| Warning | Yellow caution card for recoverable states (e.g. a task stopped because an iteration budget was exceeded, partial completion, retryable tool failure) — not a system fault |
| Error | Red-bordered card with an error icon |
Iteration Budgets
Ora runs within configurable iteration budgets (Settings → Ora → Behavior → Specialist Iteration Budget and Orchestrator Iteration Budget). When a budget is reached:
- For a specialist sub-task, Ora shows a Continue / Pause card — choose Continue to let it keep working with a fresh budget, or Pause to stop and re-prompt.
- When the orchestrator loop is cut, Ora posts a Warning card explaining that the turn stopped after the configured number of iterations — it is a "task too big, split it" signal, not an error.
Thinking Indicator
While the agent is processing, one line above the input shows Thinking with a brain mark. It is the panel's busy signal, not a reasoning indicator: it appears before the first token, during tool calls, and while the model thinks — and it disappears the moment the reply, or the chain-of-thought row, is on screen. The active agent label/badge reflects which specialist is running.
When Settings → Ora → Behavior → Show thinking is off, the model's reasoning is captured into the conversation but nothing expandable is shown. With it on, each reasoning block gets a row above its reply: the mark while the thought streams, a one-line preview of the newest thought, and — once that thought completes — Thought for Ns. Expand the row to read the whole thought.
Context Window & Token Ring
A token ring in the input status bar shows estimated context usage. When usage gets high a context pressure banner appears; you can Compact now to summarize older turns, or Dismiss it. When the conversation is near the context limit, sending is disabled until you compact or start a new conversation.
Model Selection
A model picker in the input area lets you choose the AI model for the current conversation. Available models come from the LLM profiles configured in Settings → Ora.
- When a single profile is configured, the picker is a flat list of that profile's models.
- When more than one profile is configured, it becomes a two-level flyout (provider → model). The active profile is selected implicitly by choosing its model.
- Models that support extended thinking also show a Thinking effort selector with levels Off, Low, Medium, High, and Maximum.
Detach / Attach
Click Open in window (detach button) to move Ora to a separate window. In the detached window, use the attach control to return it to the sidebar.
Interaction Style
Ora's tone is configurable in Settings → Ora → Interaction Style. Three styles are available:
| Style | Behavior |
|---|---|
| Guided | Step-by-step explanations with full context — best when exploring the product or learning the "why". |
| Direct | Concise and to the point; assumes technical familiarity and skips preamble. |
| Analyst | Business-focused and risk-aware; explains outcomes in domain terms and highlights edge cases and coverage. |
See Also
- Getting Started with Ora
- Using AI for Manual Testing
- Building Tests with AI
- Debugging & Diagnosis with AI
- Ora settings — configuring profiles, providers, interaction style, and MCP servers