> ## Documentation Index
> Fetch the complete documentation index at: https://docs.intellixent.ai/llms.txt
> Use this file to discover all available pages before exploring further.
# Configuration best practices
> A practical checklist for fine-tuning your assistant's mode, transcriber, model, and other settings for the best call experience.
Getting great results from your AI assistant comes down to picking the right engine settings. Use this guide when configuring a new assistant or troubleshooting an existing one.
## 1. Pick a mode
| Mode | Why choose it | Recommended model |
| -------------------- | ------------------------------------------------ | -------------------------------------- |
| **Dualplex (Beta)** | Fast turn-taking + premium or cloned voices | Gemini Flash 2.0/2.5 or GPT-5 Realtime |
| **Speech-to-speech** | Fastest turn-taking and most natural flow | GPT-5 Realtime |
| **Pipeline** | Maximum control over voice and long-form replies | GPT-5 Mini |
Experiment with all three modes. Record the same scenario in each and compare response time and caller satisfaction before committing to one.
See [assistant modes](/ai-assistants/assistant-modes) for a full comparison.
## 2. Choose a transcriber (Pipeline only)
| Transcriber | Accuracy | Latency | Best for |
| ------------ | -------- | ------- | ---------------------------------------------------- |
| **Azure** | ⭐⭐⭐⭐ | Slower | Highest transcription fidelity |
| **Gladia** | ⭐⭐⭐ | Faster | Good all-rounder for most languages |
| **Deepgram** | ⭐⭐⭐ | Faster | Solid choice — test against Gladia for your language |
Different languages, accents, and background noise affect each transcriber differently. Run a quick A/B test and keep the better performer.
## 3. Select an LLM model
| Model | Strengths | Trade-offs |
| ------------------------ | --------------------------------------------- | -------------------------------------------------------- |
| **GPT-5 Mini** | Balanced reasoning with low latency | May be slower than realtime models for rapid turn-taking |
| **GPT-5 Realtime** | Designed for ultra-low-latency voice turns | Best for speech-to-speech and Dualplex |
| **GPT-4o** | Strong reasoning and multimodal understanding | Higher latency |
| **Gemini Flash 2.0/2.5** | Ultra-fast for voice turns | Excellent for Dualplex and multimodal |
If speed is critical, use **GPT-5 Realtime** (great for speech-to-speech) or **Gemini Flash 2.0/2.5** (great with Dualplex). For richer reasoning, use **GPT-4o** or **GPT-5 Mini** and offset the higher latency with [filler audio](/ai-assistants/voice-selection).
## 4. Noise cancellation
* **Turn ON** when callers are on speakerphone or in noisy environments for cleaner transcription.
* **Turn OFF** if words are being clipped or the assistant is missing parts of what callers say.
If your assistant isn't hearing callers clearly, try disabling noise cancellation first.
## 5. Conversation timers
| Parameter | Recommended | Why |
| -------------------------- | ----------- | ------------------------------------------------------------------ |
| **Re-engagement interval** | \~30 s | Gives callers enough time to think. Lower values can feel pushy. |
| **Max silence duration** | \~60 s | Prevents premature hang-ups while still ending truly silent calls. |
Test with real calls — too low can interrupt callers, too high leaves awkward gaps.
## 6. Initial message
| Mode | How it's used | Best practice |
| -------------------- | ---------------------------------------- | ---------------------------------------------------------------------------------------------------------- |
| **Pipeline** | Read exactly as written (TTS conversion) | Write the greeting verbatim: "Hello, this is Alex from…" |
| **Dualplex** | Read exactly as written (ElevenLabs TTS) | Write verbatim, then select your cloned voice |
| **Speech-to-speech** | Interpreted as a prompt by the model | Include instructions like "Greet the customer and say…" or prepend `say exactly:` to ensure literal output |
## 7. Ambient sound
Ambient sound adds subtle background noise under the assistant's voice to mask processing delays and create a more natural audio experience. It is enabled by default.
If your assistant isn't hearing callers well, try turning off ambient sound or lowering its volume.
## 8. Endpointing sensitivity
Control when your assistant starts talking after a caller finishes speaking.
| Setting | Effect | Use when |
| ---------------------- | -------------------------------------------- | ------------------------------------------ |
| **Lower sensitivity** | Assistant responds faster after caller stops | You want snappy, quick-turn conversations |
| **Higher sensitivity** | Assistant waits longer before responding | Callers give longer, more detailed replies |
If your assistant cuts off callers mid-sentence, increase sensitivity. If responses feel sluggish, decrease it.
## 9. Debug with the call transcript
If something isn't working as expected:
Navigate to the **Call history** page in your dashboard.
Click on the most recent call you tested.
The call transcript includes all function calls and their parameters, making it easy to spot where the assistant went off-track.
For a complete list of every toggle and slider, see the [general settings reference](/ai-assistants/assistant-configuration).