SIP Bot

This bot follows https://docs.livekit.io/agents/start/voice-ai-quickstart/.

To test outbound call, do

lk --project <project-name> dispatch create --new-room --agent-name <agent-name> --metadata '{"phone_number": "<your-personal-number>"}'

To create agent, run

lk agent create

After making some changes, run

lk agent deploy

Testing rust-sdks / krisp-plugin dylib swaps

This agent can be used as a harness to verify changes to the LiveKit FFI dylib (liblivekit_ffi.dylib) or the Krisp noise-cancellation plugin dylib (liblivekit_nc_plugin.dylib). Rebuild the dylibs locally, swap them into this venv, and run the agent either against a synthetic audio publisher (lk load-test) or against your own mic via a browser.

One-time setup

Run all commands from the sip-bot repo root. Set $WORKSPACE to the directory that holds your rust-sdks and krisp-noise-filter checkouts.

# Directory containing rust-sdks and krisp-noise-filter (change to your layout)
WORKSPACE=/path/to/workspace

# Swappable dylib paths inside this venv (relative to the sip-bot repo root)
FFI=.venv/lib/python3.13/site-packages/livekit/rtc/resources/liblivekit_ffi.dylib
NC=.venv/lib/python3.13/site-packages/livekit/plugins/noise_cancellation/resources/liblivekit_nc_plugin.dylib

# Local builds
FFI_NEW=$WORKSPACE/rust-sdks/target/release/liblivekit_ffi.dylib
NC_NEW=$WORKSPACE/krisp-noise-filter/packages/rust/target/release/liblivekit_nc_plugin.dylib

# Back up the shipped dylibs once
[ -f "$FFI.orig" ] || cp "$FFI" "$FFI.orig"
[ -f "$NC.orig"  ] || cp "$NC"  "$NC.orig"

# Load project creds for lk + the agent
set -a; source .env.local; set +a

Build the dylibs before running any of the variants below:

(cd "$WORKSPACE/rust-sdks" && cargo build --release -p livekit-ffi)
(cd "$WORKSPACE/krisp-noise-filter/packages/rust" && cargo build --release -p krisp-ffi-plugin)

Swap dylibs for the run you want

# shipped FFI + new plugin
cp "$FFI.orig" "$FFI" && cp "$NC_NEW"  "$NC"

# new FFI + new plugin
cp "$FFI_NEW"  "$FFI" && cp "$NC_NEW"  "$NC"

# new FFI + shipped plugin
cp "$FFI_NEW"  "$FFI" && cp "$NC.orig" "$NC"

Option A — Automated with lk load-test (synthetic publisher)

Agent joins a named room; lk load-test publishes a synthetic audio track into the same room. Good for quickly confirming the filter dispatch path.

pkill -9 -f "agent.py connect" 2>/dev/null; sleep 2
ROOM=test-room-auto
rm -f /tmp/run.log
nohup uv run python agent.py connect --room "$ROOM" > /tmp/run.log 2>&1 &
until grep -q "input stream attached" /tmp/run.log; do sleep 2; done

lk load-test \
  --url "$LIVEKIT_URL" --api-key "$LIVEKIT_API_KEY" --api-secret "$LIVEKIT_API_SECRET" \
  --room "$ROOM" --audio-publishers 1 --duration 15s

grep -iE "krisp|rust-sdks" /tmp/run.log

Option B — Manual with your mic via LiveKit Meet

Agent joins a named room; you join the same room from your browser via meet.livekit.io with mic enabled. Good for listening to NC quality directly.

pkill -9 -f "agent.py connect" 2>/dev/null; sleep 2
ROOM=test-room-live
rm -f /tmp/run.log
nohup uv run python agent.py connect --room "$ROOM" > /tmp/run.log 2>&1 &
until grep -q "input stream attached" /tmp/run.log; do sleep 2; done

TOKEN=$(lk token create --identity me --room "$ROOM" --join --valid-for 1h 2>&1 \
        | awk '/^Access token:/ {print $3}')
ENCODED_URL=$(python3 -c "import urllib.parse,sys;print(urllib.parse.quote(sys.argv[1]))" "$LIVEKIT_URL")
open "https://meet.livekit.io/custom?liveKitUrl=$ENCODED_URL&token=$TOKEN"

# In another terminal, tail the log while you speak:
tail -f /tmp/run.log | grep -iE "krisp|rust-sdks"

Cleanup

pkill -9 -f "agent.py connect"
cp "$FFI.orig" "$FFI"
cp "$NC.orig"  "$NC"

Observed results

Tested with temporary eprintln! logs added on the two PR branches (rust-sdks #1019, krisp-noise-filter #174) to make the dispatch path visible at runtime:

• rust-sdks livekit/src/plugin.rs::new_session — prints [rust-sdks] audio_filter v2 dispatch: {in}->{out} Hz or [rust-sdks] audio_filter v1 dispatch: {rate} Hz depending on which branch is taken.
• krisp audio_filter_create / audio_filter_create_v2 — prints [krisp-plugin] audio_filter_create (v1): {rate} Hz and/or [krisp-plugin] audio_filter_create_v2: {in}->{out} Hz.

Run via Option A above on macOS arm64, Opus publishers at 48 kHz, agent configured for 24 kHz output (the livekit-agents default).

Scenario	$FFI	$NC	Observed stderr	Verdict
Old SDK + new plugin (backwards-compat)	shipped (v1)	new (#174)	[krisp-plugin] audio_filter_create (v1): 24000 Hz → [krisp-plugin] audio_filter_create_v2: 24000->24000 Hz	Old SDKs keep working with the new plugin. The v1 entrypoint correctly delegates to v2 with equal rates.
New SDK + new plugin (v2 fast path)	new (#1019)	new (#174)	[rust-sdks] audio_filter v2 dispatch: 48000->24000 Hz → [krisp-plugin] audio_filter_create_v2: 48000->24000 Hz	v2 probe + dispatch fires; plugin v2 entrypoint receives asymmetric rates; WebRTC sink is at 48 kHz (passthrough), filter does 48→model_rate→24.
New SDK + old plugin (v1 fallback)	new (#1019)	shipped (v1 only)	[rust-sdks] audio_filter v1 dispatch: 24000 Hz	v2-capable SDK falls back cleanly when the plugin has no _v2 symbols. No regression for v1-only plugins (ai-coustics etc.).

The eprintlns are development-only — revert them before merging either PR.