> ## Documentation Index
> Fetch the complete documentation index at: https://docs.autocalls.ai/llms.txt
> Use this file to discover all available pages before exploring further.
# Assistant best practices
> Quick guide to fine-tune mode, transcriber, model, and other settings for the best call experience.
> **Last updated:** September 1, 2025
Getting great results often comes down to **picking the right engine settings**. Use this checklist when configuring an assistant:
## 1. Pick a Mode
| Mode | Why choose it? | Notes |
| --------------------------------- | ---------------------------------------------- | ------------------------------------------------------------------------------------------------------ |
| **Dualplex (Beta)** | Fast turn-taking + premium/cloned voices | Recommended default. Pair with **Gemini Flash 2.0/2.5** or **GPT‑5 Realtime**. |
| **Speech-to-Speech (Multimodal)** | Fastest turn-taking & most natural flow | Recommended model: **GPT‑5 Realtime**. |
| **Pipeline** | Maximum control over voice & long-form replies | Recommended model: **GPT‑5 Mini**. If you select Pipeline, continue to the **Transcriber** step below. |
Want to know more about the differences between the modes? Read the Assistant modes guide.
Experiment with all three modes: record the same scenario in each and compare response time and caller satisfaction.
## 2. Choose a Transcriber (Pipeline only)
| Transcriber | Accuracy | Latency | Best for |
| ------------ | -------- | --------------- | -------------------------------------------------------------------------------- |
| **Azure** | ⭐⭐⭐⭐ | ⏱️⏱️⏱️ (slower) | When you need the highest transcription fidelity. |
| **Gladia** | ⭐⭐⭐ | ⏱️ (faster) | Good all-rounder for most languages. |
| **Deepgram** | ⭐⭐⭐ | ⏱️ (faster) | Another solid choice—test which performs better for your language & audio setup. |
> **Tip:** Different languages, accents, or background noise can impact each engine differently. Run a quick A/B test and keep the best 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 in Dualplex/Multimodal | Excellent for minimizing perceived latency. |
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 latency by using filler audios.
## 4. Noise Cancellation
If callers are on speaker phone or in a quiet environment, keep **noise cancellation ON**. If your call volume is low or some words are "clipped," **turn it OFF** so the transcriber gets the full waveform.
If your assistant is not hearing you well, you can try to turn off noise cancellation.
## 5. Conversation Timers
| Parameter | Recommended | Why |
| ------------------------ | ----------- | ------------------------------------------------------------------ |
| **Re-engagement** | `≈ 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 different values in real calls—too low can interrupt, too high leaves awkward gaps.
## 6. Initial Message
| Mode | How it's used | Best practice |
| -------------------- | ---------------------------------------------------------- | --------------------------------------------------------------------------------------------------------------- |
| **Pipeline** | Read **exactly** as written (converted by TTS). | Write the greeting verbatim: "Hello, this is Alex from …". |
| **Dualplex** | Read **exactly** as written (rendered via ElevenLabs TTS). | Write the greeting 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
Enabled by default, ambient sound is a feature that adds background noise to the assistant's voice.
If the assistant is not hearing you well, you can try to turn off ambient sound or turn the volume lower.
## 8. Endpointing sliders
**Control when your assistant starts talking** with the endpointing sensitivity slider at the bottom of assistant settings.
| Setting | Effect | Use when |
| ---------------------- | --------------------------------------------------------- | ------------------------------------------ |
| **Lower sensitivity** | Assistant responds **faster** after caller stops speaking | You want snappy, quick-turn conversations |
| **Higher sensitivity** | Assistant waits **longer** before responding | Callers give longer, more detailed replies |
**Pro tip:** If your assistant cuts off callers mid-sentence, **increase** the sensitivity. If responses feel sluggish, **decrease** it.
## 9. Debug using call transcript
If you are having issues with your assistant, you can use the call transcript to debug the issue.
1. Go to the Call history page.
2. Click on the last call you tested
3. The call transcript will be shown including function calls and its parameters.
## 10. Still have questions?
If you have any questions, please contact our support team via the chat widget inside the app.
Test different settings with real calls—the right balance depends on your conversation flow and caller behavior patterns.
***
Need a complete list of every toggle and slider? See the full
Assistant settings reference.