Do I need to tune every advanced setting manually?

No. In roughly 90% of cases, the default configuration is already the best practice, and you only need to change one to three clearly relevant items.

How can I control the final answer language separately?

Use the extra instruction field in answer preferences and state the output language directly, such as asking the model to always reply in Chinese.

Which settings should I check first when results feel unstable?

Start with answer preferences, capture-related settings, voice settings, and shortcut conflicts instead of changing many parameters at once.

Settings and Advanced Settings

This page explains desktop settings and advanced controls so you can tune Interview AiBox for your interview workflow. If you want more scenario-driven guidance on how to combine these settings, continue with Advanced Usage.

Bottom line first

In roughly 90% of cases, the default setup is already the best-practice setup. Our defaults cover most real interview, coding test, and common screen-sharing workflows. Change settings only when you have a clear reason, such as coding format, reply language, screenshot mode, hotwords, or platform-specific compatibility.

Basic Settings

Basic settings are available on the main settings panel:

Theme: light / dark
Programming Language: default language for coding answers
Interview Language: response language preference
Problem-solving Model / Voice Model: choose defaults; voice can also start with Smart route when you do not want to pin one route manually
App Mode / Coding format preference: general / LeetCode / ACM. Keep General for non-coding interviews; LeetCode favors function-style output, while ACM favors a full runnable program
Screenshot Mode: full screen / region
Voice Mode: standard / flagship
Interview Planning: load cloud plans and linked resume

A smooth first run usually depends on getting the basics right before touching advanced settings, especially coding format preference, interview language, screenshot mode, and interview planning.

When Interview Planning is enabled, company/role/language can be auto-filled and resume linking is supported.

Advanced Settings

Advanced settings are not meant to be tuned aggressively. The defaults are already optimized for the common interview workflow, so most users only need to change 1-3 clearly relevant items.

1) Interface Opacity

Adjust toolbar and panel background opacity to reduce visual obstruction.

2) Hotkey Customization

Configure hotkeys per page. Press Enter to save each item, or reset to defaults in one click. The default hotkeys already cover the main workflow, so only change them when they conflict with your OS, input method, or another app.

3) Answer Preferences

Answer style: concise, bullet-based, conclusion first, reasoning first, etc.
Additional instructions: stable prompt-level guidance
Coding format preference: LeetCode works better for function-only answers; ACM works better for full main plus standard input/output; keep General for non-coding interviews
Auto-submit after screenshot: switch into screenshot Q&A and submit the current screenshot automatically, reducing one extra hotkey action
Streamed answer rendering: show response generation in real time

Common writing ideas:

Translation scenarios: "Keep technical terms in English and give natural spoken translation instead of literal translation."
English interviews for non-native speakers: "Use concise spoken English, start with the conclusion, then add 1-2 supporting points."
Behavioral rounds: "Use background-action-result structure, keep it within 4 sentences, include one metric if possible."
Coding rounds: "Explain approach and complexity first, then code in the required format."
Screenshot Q&A: this is uncommon, but if full-screen capture makes the prompt hard to read because resolution is too low, the text is too dense, or recognition is not precise enough, switch to region capture first.

4) Custom Models

Support OpenAI-compatible APIs:

Display name / Base URL / API key / model name
Extra request headers and JSON body fields
Multimodal switch

Custom model credentials are local-only and never synced. Some features may require higher-tier plans.

5) Cloud Configuration

Session cloud sync: uploads transcripts and session metadata only, without audio or screenshots
AI review scoring: transcript-based review scoring for post-interview reflection
Cloud interview plan: load plan data and auto-fill company, role, and language
Cloud resume workflow: strongly recommended to complete the cloud resume in the dashboard first, then sync it into the knowledge base for more stable recall

6) Voice Settings

Smart route: recommended when you do not want to manually pin one speech route
Additional instructions for voice answer style, such as "give a speakable outline only" or "answer in short spoken English"
Hotwords to improve recognition of company names, project names, acronyms, and technical terms
Sentence split timeout for silence (250-2000ms)
Carry previous QA for next-turn context
Use voice knowledge base in voice answers
Message filter: interviewer-only / candidate-only / all
Recent message count: number of lines sent for analysis

Voice usage guidance:

In most cases, keep the default best-practice setup first and start with Smart route instead of changing many voice parameters at once.
Formal interview rounds usually work best with Interviewer only when the interviewer is doing most of the talking.
If you want fewer actions, the system is better positioned to auto-trigger Q&A after a complete interviewer question; if the round is long, multi-speaker, or interruption-heavy, manual timing is usually safer.
Test Smart route, hotwords, silence threshold, and message filter on the target platform or /tools before the real interview.

7) Knowledge Base Settings

Intelligent chunking mode: preserve project boundaries during indexing
Context extension: fetch neighboring chunks during recall
topK parameter: number of recalled chunks (resume 7, QA 5, technical doc 7)

The knowledge base panel also works as a stealth-friendly note layer. You can keep it open alongside voice and screenshot Q&A instead of treating it as a separate mode.

Knowledge base retrieval settings also should not be tuned frequently. The default chunking, context extension, and topK values already cover most resume, project, and technical-document workflows. Adjust them only when you have a clear recall problem.