# 01 Let me finish

> Deepgram workshops › Build a voice agent › Mission 01
> Web version: https://workshops.deepgram.com/voice-agent/let-me-finish
> About 25 min (estimate). Check it: `make check 01` (offline, no key).

Read a phone number slowly and it waits.

- [The knobs](https://workshops.deepgram.com/voice-agent/let-me-finish/the-knobs.md)
- [Read a number](https://workshops.deepgram.com/voice-agent/let-me-finish/read-a-number.md)
- [Change one thing](https://workshops.deepgram.com/voice-agent/let-me-finish/change-one-thing.md)

## The knobs

This mission needs early access to the `timing` feature for live runs. Its offline check works without it.

Read a phone number with pauses and the agent waits for the rest. Three knobs set how:

| Knob | What it sets |
|---|---|
| **Backchannel sensitivity** | How readily it says "mm-hm" |
| **Presence check interval** | Silence before "still there?" |
| **Max presence checks** | How many times it asks |

**How it works**

A voice agent that goes on silence alone treats a long pause as the end of your turn. Read a number the way you would to a person, three digits, a breath, four more, and it answers in the breath. Deepgram Flux STT (`flux-general-en`) detects the end of your turn, and the pause is measured from there. With {{name.timing}} on, the agent listens to how you paused as well as when, and decides whether to wait.

**Backchannel sensitivity:** at 0 it never interjects and every pause is yours to fill; near 1 it waits longer and reassures more. **Presence check interval:** short feels pushy, long leaves a dropped call open. **Max presence checks:** 0 means it never asks. The agent stays on the line after the last check until you or your app hang up.

Settings open with the voice you picked in First words; pick it again if not. The prompt is a receptionist taking a number, since a phone number is the shortest thing people say with pauses in it.

## Read a number

### Read your number slowly

**Do:** Tap **Start the agent**, say "I'd like to leave my number," then read seven digits out loud with a real pause after the first three.

**Listen for:** "mm-hm" in the gap.

_On the web page, this step has the agent console: talk or type to the agent there, or run the same mission in your own code._

**Check:** It waited while you read a phone number.

### Say goodbye

**Do:** When it has your number, say "okay, bye" after it finishes a sentence.

**Listen for:** its goodbye; the call closes on its own.

**Check:** It wrapped up when you said goodbye.

**How it works**

Backchannel sensitivity starts at 0.3, enough for a short pause; leave the presence knobs for now. Talk out loud here: the agent hears the pause in your voice.

A quiet "mm-hm" in the gap means it's holding the floor. An answer means it took the turn.

The agent never closes the connection itself. It calls a function named `end_call`, which this mission gives it, and your browser hangs up once the goodbye has played. The status changes to Conversation ended with no tap on **Hang up**.

## Change one thing

### Set sensitivity to zero

**Do:** Set **Backchannel sensitivity** to 0, tap **Start the agent again**, and read the same number.

**Listen for:** who takes the gap. Set it back to 0.3 and tap **Reconnect and apply**.

_On the web page, this step has the Settings panel; in your own code the same settings are in the start file._

_On the web page, this step shows the code panel with the matching code._

_On the web page, this step has the agent console: talk or type to the agent there, or run the same mission in your own code._

**Check:** You heard what Backchannel sensitivity does. (Optional.)

### Try the presence knobs

**Do:** Stay quiet. Leave the presence knobs where they are.

**Listen for:** "still there?" after about 4 seconds, twice, then silence.

### Hard mode: ten digits, two pauses

**Do:** **Hard mode.** Read ten digits aloud in three pieces.

**Listen for:** it waiting through both pauses. Then tap **Hang up**.

**Check:** Hard mode: ten digits, two pauses. (Optional.)

### Take it home

Take it home: [Python](https://github.com/deepgram-devs/voice-agent-workshop/blob/main/python/missions/01-let-me-finish/start.py) or [TypeScript](https://github.com/deepgram-devs/voice-agent-workshop/blob/main/typescript/missions/01-let-me-finish/start.ts).

**How it works**

Say what changed in one sentence before you mark it: "at zero it answered after three digits; at 0.3 it hummed and waited."

The three sliders are three numbers in the timing block of the Settings message; the **Settings JSON** tab shows your values. The handlers in the code panel show where "mm-hm" and the presence check arrive, and how your app hangs up when the agent calls `end_call`.

The silence you counted is **Presence check interval**; the number of asks is **Max presence checks**. A max of 0 never asks; a short interval with a max of 5 asks five times close together. Tap **Hang up** when you're done listening. Change either one and tap **Reconnect and apply** to send it.

**Hard mode:** three digits, a pause, three more, a pause, the last four. If the call has ended, tap **Start the agent again** first. The check counts the pieces your voice arrived in and passes when the agent let you finish all ten. If it jumps in, raise **Backchannel sensitivity** a notch and tap **Reconnect and apply**.

The attendee project has this mission as four TODOs for the timing block and `end_call`, with an offline check.

The what-changed check is optional and on your honor.

**In production**

- **Watch:** talk-overs. A `UserStartedSpeaking` within a second or two of `AgentStartedSpeaking` usually means the agent took a turn the caller wasn't done with. Count them per call.
- **Aim for:** Backchannel sensitivity between 0.3 and 0.6 for callers reading numbers and addresses, and a presence check every 4 to 8 seconds with 1 or 2 checks before your app hangs up.
- **Read:** the [Mission 01 brief](https://github.com/deepgram-devs/voice-agent-workshop/blob/main/missions/01-let-me-finish/README.md) in the code project: each knob, what it does, and its wire name.

**What you tuned:** the timing layer. Same prompt, same voice, and now it knows when you're done talking.
