rtpbench 0.3.0__py3-none-any.whl

rtpbench/targets/aivc/aivc.ini

@@ -0,0 +1,25 @@
+# AI Voice Connector CE settings for rtpbench: the openai flavor only, as upstream's example sets
+# it (voice, instructions, welcome message), plus a transfer target so the refer cases can run:
+# the operator (rtpbench's transfer target, compose.yaml).
+# The key and URL come from the environment (OPENAI_API_KEY, OPENAI_URL).
+[opensips]
+ip = 127.0.0.1
+port = 8080
+
+[engine]
+event_ip = 127.0.0.1
+
+[openai]
+voice = alloy
+instructions = You are a helpful assistant.
+welcome_message = Hello! How can I help you?
+transfer_to = sip:operator@operator
+
+[deepgram]
+disabled = true
+
+[deepgram_native]
+disabled = true
+
+[azure]
+disabled = true

rtpbench/targets/aivc/compose.yaml

@@ -0,0 +1,72 @@
+# OpenSIPS AI Voice Connector CE (openai flavor) as a target, merged over ../../compose.yaml.
+# `target` owns the network namespace (agentnet, plus egress to the AI engine); OpenSIPS and the
+# connector share it, as upstream's example shares the host's. rtpbench routes the caller via
+# the router, so SIP and RTP cross netem. Needs OPENAI_API_KEY (and OPENAI_URL for a gateway).
+# Paths are relative to the first compose file's directory.
+services:
+  target:
+    image: ${RTPBENCH_IMAGE:?}
+    cap_add: [NET_ADMIN]
+    init: true
+    depends_on: [router]
+    networks: [agentnet, egress]
+    command: sleep infinity
+
+  opensips:
+    image: opensips/opensips:3.6@sha256:81a7fa3c7bdbceaadb7cb869ff7ac3f34efb2d95cd7fecb3b689d7b6b1eb0672
+    network_mode: service:target
+    depends_on: [target]
+    volumes: ["./targets/aivc/opensips.cfg:/etc/opensips/opensips.cfg:ro"]
+    healthcheck:
+      test: opensips-cli -x mi sr_get_status core | grep -q running
+      interval: 1s
+      retries: 20
+
+  engine:
+    build:
+      context: https://github.com/OpenSIPS/opensips-ai-voice-connector-ce.git#b95f5afa99301e5d6ee0643dcea73b1a457b6286
+      # upstream's docker/Dockerfile, with the base image and every installed package pinned
+      # (upstream pins only its direct requirements)
+      additional_contexts: { pins: ./targets/aivc }
+      dockerfile_inline: |
+        FROM python:3.14-trixie@sha256:446efb4225200a0a196707aa7953898b37962fd3b8859a636035de3af657081e
+        WORKDIR /app
+        COPY . /app/
+        COPY --from=pins constraints.txt /tmp/constraints.txt
+        RUN pip install --no-cache-dir -r requirements.txt -c /tmp/constraints.txt
+        CMD ["python3", "-u", "src/main.py"]
+    image: ${RTPBENCH_REGISTRY:?}/rtpbench-aivc:b95f5af${RTPBENCH_BUILD_ENGINE:-}
+    network_mode: service:target
+    depends_on:
+      opensips: { condition: service_healthy }
+    volumes: ["./targets/aivc/aivc.ini:/app/cfg/aivc.ini:ro"]
+    environment:
+      CONFIG_FILE: /app/cfg/aivc.ini
+      OPENAI_API_KEY: ${OPENAI_API_KEY:?set OPENAI_API_KEY}
+      OPENAI_URL: ${OPENAI_URL:-wss://api.openai.com/v1/realtime?model=gpt-realtime-2.1}
+      # upstream's default; empty for a model without reasoning (gpt-realtime-1.5), as its docs say
+      OPENAI_REASONING_EFFORT: ${OPENAI_REASONING_EFFORT-low}
+    # ready once OpenSIPS lists it as the subscriber for call events
+    healthcheck:
+      test:
+        - CMD
+        - python3
+        - -c
+        - >-
+          import json, socket; s = socket.socket(2, 2); s.settimeout(1);
+          s.sendto(json.dumps({"jsonrpc": "2.0", "id": 1, "method": "subscribers_list",
+          "params": ["E_UA_SESSION"]}).encode(), ("127.0.0.1", 8080));
+          assert json.loads(s.recv(65535))["result"]["Event"]["Subscribers"]
+      interval: 1s
+      retries: 30
+    # SDP carries the address facing the caller (the router's side), not the egress one
+    command:
+      - sh
+      - -c
+      - >-
+        RTP_IP=$$(python3 -c 'import socket as s; u = s.socket(2, 2);
+        u.connect((s.gethostbyname("router"), 9)); print(u.getsockname()[0])')
+        exec python3 -u src/main.py
+
+networks:
+  egress:

rtpbench/targets/aivc/constraints.txt

@@ -0,0 +1,52 @@
+aioice==0.10.2
+aiortc==1.15.0
+annotated-types==0.8.0
+anyio==4.15.1
+attrs==26.1.0
+av==17.1.0
+azure-cognitiveservices-speech==1.51.2
+azure-core==1.41.0
+certifi==2026.7.22
+cffi==2.1.1
+charset-normalizer==3.5.2
+click==8.5.0
+cryptography==50.0.2
+deepgram-sdk==7.8.1
+dnspython==2.9.0
+google-crc32c==1.9.0
+h11==0.16.0
+httpcore==1.0.9
+httpcore2==2.13.1
+httpx==0.28.1
+httpx2==2.13.1
+idna==3.20
+ifaddr==0.2.0
+jiter==0.17.0
+jsonschema==4.26.0
+jsonschema-specifications==2025.9.1
+mcp==2.3.0
+mcp-types==2.3.0
+openai==3.11.0
+opensips==0.1.10
+opentelemetry-api==1.45.1
+pycparser==3.11
+pydantic==2.14.1
+pydantic_core==2.50.1
+pyee==14.0.0
+PyJWT==2.15.1
+pylibsrtp==1.0.0
+pyOpenSSL==26.4.0
+python-multipart==0.0.32
+referencing==0.37.0
+requests==2.34.2
+rpds-py==2026.9.1
+sipmessage==0.7.1
+sniffio==1.3.1
+sse-starlette==3.5.0
+starlette==1.8.0
+truststore==0.10.5
+typing-inspection==0.4.4
+typing_extensions==4.16.0
+urllib3==2.8.0
+uvicorn==0.54.0
+websockets==17.1

rtpbench/targets/aivc/opensips.cfg

@@ -0,0 +1,45 @@
+# SPDX-License-Identifier: GPL-3.0-or-later
+# From OpenSIPS AI Voice Connector CE (c) SIP Point Consulting SRL, examples/simple/b2b/opensips.cfg
+# at b95f5af, with its variables filled in (SIP on 5060, MI on 8080). Under the GPL like upstream.
+
+log_level=3
+xlog_level=2
+stderror_enabled=yes
+syslog_enabled=no
+
+socket=udp:*:5060
+socket=tcp:*:5060
+
+mpath="/usr/lib/x86_64-linux-gnu/opensips/modules/"
+
+loadmodule "tm.so"
+loadmodule "rr.so"
+loadmodule "sipmsgops.so"
+loadmodule "signaling.so"
+loadmodule "cfgutils.so"
+loadmodule "mi_fifo.so"
+loadmodule "proto_tcp.so"
+loadmodule "proto_udp.so"
+loadmodule "sl.so"
+loadmodule "event_datagram.so"
+loadmodule "mi_datagram.so"
+loadmodule "b2b_entities.so"
+
+modparam("mi_fifo", "fifo_name", "/tmp/opensips_fifo")
+
+modparam("tm", "fr_timeout", 2)
+modparam("tm", "fr_inv_timeout", 3)
+modparam("tm", "restart_fr_on_each_reply", 0)
+modparam("tm", "onreply_avp_mode", 1)
+
+modparam("mi_datagram", "socket_name", "udp:127.0.0.1:8080")
+
+route {
+	if (!is_method("INVITE") || has_totag()) {
+		send_reply(405, "Method not allowed!\n");
+		exit;
+	}
+	ua_session_server_init($var(b2b_key), "drbh");
+	xlog("Started new call for $var(b2b_key)\n");
+	exit;
+}

rtpbench/targets/ava/ai-agent.yaml

@@ -0,0 +1,275 @@
+# AVA's golden OpenAI Realtime configuration (config/ai-agent.golden-openai.yaml at
+# ba52df92bc1cad2c1f277034e1423b80c8f50be4), with rtpbench's edits marked "rtpbench:".
+#
+# Copyright (c) 2025 Asterisk AI Voice Agent contributors
+#
+# Permission is hereby granted, free of charge, to any person obtaining a copy of this software
+# and associated documentation files (the "Software"), to deal in the Software without
+# restriction, including without limitation the rights to use, copy, modify, merge, publish,
+# distribute, sublicense, and/or sell copies of the Software, and to permit persons to whom the
+# Software is furnished to do so, subject to the following conditions:
+#
+# The above copyright notice and this permission notice shall be included in all copies or
+# substantial portions of the Software.
+#
+# THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR IMPLIED, INCLUDING
+# BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+# NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM,
+# DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+# OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+
+# ═══════════════════════════════════════════════════════════════════════════
+# Golden Baseline #1: OpenAI Realtime - Cloud Monolithic Agent
+# ═══════════════════════════════════════════════════════════════════════════
+#
+# VALIDATED CONFIGURATION - Production-ready
+#
+# Requirements:
+#   - OPENAI_API_KEY in .env file
+#   - Asterisk 18+ with ARI and AudioSocket modules
+#
+# Performance:
+#   - Response time: <2 seconds (typically 0.5-1.5s)
+#   - Audio quality: Excellent (native 24kHz output)
+#   - Cost: ~$0.06 per minute (OpenAI Realtime pricing)
+#
+# Best for:
+#   - Quick setup and deployment
+#   - Modern, natural-sounding conversations
+#   - Enterprise deployments requiring reliability
+#
+# For detailed parameter explanations, see: docs/Configuration-Reference.md
+# ═══════════════════════════════════════════════════════════════════════════
+
+# Active Configuration
+config_version: 6
+active_pipeline: local_hybrid  # Don't change - uses provider override via contexts
+default_provider: openai_realtime
+
+# Audio Transport - AudioSocket for OpenAI Realtime
+# rtpbench: AudioSocket as shipped, ExternalMedia (RTP) or WebSocket, from
+# targets/ava/compose.yaml
+audio_transport: ${AVA_TRANSPORT}
+audiosocket:
+  format: slin
+  host: 0.0.0.0
+  port: 8090
+# rtpbench: WebSocket's listener, as docs/WebSocket-Transport.md configures it on one host,
+# matching targets/ava/websocket_client.conf
+websocket_media:
+  connection_mode: asterisk_outbound
+  connection_name: aava_media
+  bind_host: 127.0.0.1
+  advertise_host: 127.0.0.1
+  port: 8787
+  path: /media
+  format_policy: profile
+  fallback_format: ulaw
+  control_format: json
+  direction: both
+  allowed_remote_hosts: [127.0.0.1]
+  auth:
+    required: true
+    username: aava_media
+    password_env: ASTERISK_MEDIA_WS_PASSWORD
+  tls:
+    enabled: false
+
+# Playback Mode
+downstream_mode: stream
+
+# Contexts - Use demo_openai context to activate OpenAI Realtime
+contexts:
+  default:
+    greeting: Hello, I am a Voice Assistant. How can I help you today?
+    profile: openai_realtime_24k_${AVA_TRANSPORT}  # rtpbench: the transport's profile below
+    prompt: You are a concise voice assistant. Respond clearly and keep answers under 20 words unless more detail is requested.
+    provider: openai_realtime
+    # rtpbench: the transfer tool, to the destination below
+    tools:
+      - blind_transfer
+  demo_openai:
+    greeting: "Hi! I'm Ava running on OpenAI Realtime API. I'm fast and natural-sounding. What would you like to know about this project?"
+    prompt: "You are Ava (Asterisk Voice Agent) demonstrating OpenAI Realtime. Be helpful, conversational, and explain features clearly in 5-10 sentences."
+    profile: openai_realtime_24k_${AVA_TRANSPORT}  # rtpbench: the transport's profile below
+    provider: openai_realtime
+
+# Barge-In - Enable for natural interruption
+barge_in:
+  enabled: true
+  energy_threshold: 700
+  initial_protection_ms: 100
+  min_ms: 150
+  post_tts_end_protection_ms: 100
+
+# VAD Configuration
+vad:
+  enhanced_enabled: true
+  fallback_buffer_size: 128000
+  fallback_enabled: true
+  fallback_interval_ms: 4000
+  max_utterance_duration_ms: 10000
+  min_utterance_duration_ms: 600
+  use_provider_vad: false
+  utterance_padding_ms: 200
+  webrtc_aggressiveness: 1
+  webrtc_end_silence_frames: 50
+  webrtc_start_frames: 3
+
+# Streaming Configuration
+streaming:
+  chunk_size_ms: 20
+  connection_timeout_ms: 120000
+  continuous_stream: true
+  empty_backoff_ticks_max: 5
+  fallback_timeout_ms: 8000
+  greeting_min_start_ms: 40
+  jitter_buffer_ms: 950
+  keepalive_interval_ms: 5000
+  low_watermark_ms: 80
+  min_start_ms: 120
+  normalizer:
+    enabled: true
+    max_gain_db: 18.0
+    target_rms: 1400
+  provider_grace_ms: 500
+  sample_rate: 8000
+
+# Audio Profiles - openai_realtime_24k for native 24kHz audio
+profiles:
+  default: telephony_responsive
+  # rtpbench: the golden openai_realtime_24k, as shipped for AudioSocket
+  openai_realtime_24k_audiosocket:
+    chunk_ms: 20
+    idle_cutoff_ms: 0
+    internal_rate_hz: 24000
+    provider_pref:
+      input_encoding: pcm16
+      input_sample_rate_hz: 24000
+      output_encoding: pcm16
+      output_sample_rate_hz: 24000
+    transport_out:
+      encoding: slin
+      sample_rate_hz: 8000
+  # rtpbench: the same for WebSocket, which takes its wire format (slin) from the profile
+  openai_realtime_24k_websocket:
+    chunk_ms: 20
+    idle_cutoff_ms: 0
+    internal_rate_hz: 24000
+    provider_pref:
+      input_encoding: pcm16
+      input_sample_rate_hz: 24000
+      output_encoding: pcm16
+      output_sample_rate_hz: 24000
+    transport_out:
+      encoding: slin
+      sample_rate_hz: 8000
+  # rtpbench: the same for ExternalMedia, but ulaw: slin goes out as PCMU, and the caller hears
+  # noise
+  openai_realtime_24k_externalmedia:
+    chunk_ms: 20
+    idle_cutoff_ms: 0
+    internal_rate_hz: 24000
+    provider_pref:
+      input_encoding: pcm16
+      input_sample_rate_hz: 24000
+      output_encoding: pcm16
+      output_sample_rate_hz: 24000
+    transport_out:
+      encoding: ulaw
+      sample_rate_hz: 8000
+  telephony_responsive:
+    chunk_ms: auto
+    idle_cutoff_ms: 600
+    internal_rate_hz: 8000
+    provider_pref:
+      input_encoding: mulaw
+      input_sample_rate_hz: 8000
+      output_encoding: mulaw
+      output_sample_rate_hz: 8000
+    transport_out:
+      encoding: slin
+      sample_rate_hz: 8000
+
+# Provider Configuration - OpenAI Realtime
+providers:
+  openai_realtime:
+    # rtpbench: base_url, model and the pacer from targets/ava/compose.yaml
+    base_url: ${OPENAI_BASE_URL}
+    egress_pacer_enabled: ${AVA_EGRESS_PACER}
+    egress_pacer_warmup_ms: 320
+    enabled: true
+    greeting: ${OPENAI_GREETING:-Hello, how can I help you today?}
+    input_encoding: ulaw
+    input_sample_rate_hz: 8000
+    instructions: You are a concise voice assistant. Respond clearly and keep answers under 20 words unless more detail is requested.
+    api_version: ga
+    model: ${OPENAI_MODEL}
+    organization: ''
+    output_encoding: linear16
+    output_sample_rate_hz: 24000
+    provider_input_encoding: linear16
+    provider_input_sample_rate_hz: 24000
+    response_modalities:
+    - audio
+    - text
+    target_encoding: mulaw
+    target_sample_rate_hz: 8000
+    turn_detection:
+      create_response: true
+      prefix_padding_ms: 200
+      silence_duration_ms: 200
+      threshold: 0.5
+      type: server_vad
+    voice: alloy
+
+# LLM Fallback Configuration
+llm:
+  initial_greeting: Hello, how can I help you today?
+  prompt: Voice assistant. Answer in 5-8 words. Be direct. Expand only if asked.
+
+# Pipelines - Required but not used by monolithic OpenAI provider
+pipelines:
+  local_hybrid:
+    llm: openai_llm
+    options:
+      llm:
+        base_url: https://api.openai.com/v1
+        max_tokens: 150
+        model: gpt-4o-mini
+        temperature: 0.7
+      stt:
+        chunk_ms: 160
+        mode: stt
+        stream_format: pcm16_16k
+        streaming: true
+      tts:
+        format:
+          encoding: mulaw
+          sample_rate: 8000
+    stt: local_stt
+    tts: local_tts
+
+# Asterisk Configuration
+asterisk:
+  app_name: asterisk-ai-voice-agent
+
+# rtpbench: transfers go to the operator through targets/ava/extensions.conf
+tools:
+  transfer:
+    enabled: true
+    destinations:
+      operator:
+        description: An operator, a person
+        dialplan_context: transfer
+        target: operator
+        type: extension
+
+# ExternalMedia (not used by AudioSocket, but required by config schema)
+external_media:
+  codec: ulaw
+  direction: both
+  port_range: 18080:18099
+  # rtpbench: Asterisk sends to this address, so not 0.0.0.0 (it shares the engine's netns)
+  rtp_host: 127.0.0.1
+  rtp_port: 18080

rtpbench/targets/ava/ari.conf

@@ -0,0 +1,7 @@
+; ARI for AVA's engine, which shares this network namespace (http.conf: loopback only)
+[general]
+enabled=yes
+[asterisk]
+type=user
+read_only=no
+password=asterisk

rtpbench/targets/ava/compose.yaml

@@ -0,0 +1,128 @@
+# Asterisk + AVA (Asterisk AI Voice Agent) as a target, merged over ../../compose.yaml. AVA is an
+# ARI app for an existing Asterisk; its compose runs the engine on the host's network next to
+# it. Here Asterisk and the engine share the network namespace of `target` (agentnet, plus
+# egress to the AI engine), and the engine reaches ARI and the call audio (AudioSocket,
+# ExternalMedia RTP or Asterisk's WebSocket channel, AVA_TRANSPORT) on 127.0.0.1.
+# rtpbench routes the caller via the router, so SIP and RTP cross netem. Needs OPENAI_API_KEY
+# (and OPENAI_BASE_URL for another Realtime endpoint). Paths are relative to the first compose
+# file's directory.
+services:
+  target:
+    image: ${RTPBENCH_IMAGE:?}
+    cap_add: [NET_ADMIN]
+    init: true
+    depends_on: [router]
+    networks: [agentnet, egress]
+    command: sleep infinity
+
+  asterisk:
+    build:
+      context: ./targets/ava
+      # andrius/asterisk without the AudioSocket modules (menuselect leaves them out): built
+      # from the same source and ABI options (OPTIONAL_API), and added
+      dockerfile_inline: |
+        FROM debian:trixie@sha256:913f6706df59a68922d1dd08f78c2476560a8d367897200a6005b00e5f67c2d5 AS modules
+        # the Debian snapshot this base image was built from, so every package version is fixed
+        RUN sed -i -e 's|^URIs: http://deb.debian.org/debian$$|URIs: http://snapshot.debian.org/archive/debian/20261005T000000Z|' \
+            -e 's|^URIs: http://deb.debian.org/debian-security$$|URIs: http://snapshot.debian.org/archive/debian-security/20261005T000000Z|' \
+            /etc/apt/sources.list.d/debian.sources \
+            && apt-get -o Acquire::Check-Valid-Until=false update \
+            && apt-get install -y --no-install-recommends ca-certificates \
+            build-essential curl libedit-dev libjansson-dev libncurses-dev libsqlite3-dev \
+            libssl-dev libxml2-dev uuid-dev pkg-config libcurl4-openssl-dev libsrtp2-dev
+        WORKDIR /usr/src/asterisk
+        RUN curl -fsSL -o /tmp/a.tar.gz https://downloads.asterisk.org/pub/telephony/certified-asterisk/releases/asterisk-certified-22.8-cert4.tar.gz \
+            && echo "a41a7007d420119255017bbf15759a4204fb45915b1ce5f0dc7622ecaf498c34  /tmp/a.tar.gz" | sha256sum -c - \
+            && tar --strip-components=1 -xzf /tmp/a.tar.gz && rm /tmp/a.tar.gz
+        RUN ./configure --with-ssl=ssl --with-crypto --without-pjproject-bundled \
+            --without-pjproject > /dev/null && make menuselect.makeopts > /dev/null \
+            && menuselect/menuselect --disable BUILD_NATIVE --enable BETTER_BACKTRACES \
+            --enable res_audiosocket --enable chan_audiosocket --enable app_audiosocket \
+            menuselect.makeopts
+        RUN make -j"$$(nproc)" res channels apps > /dev/null \
+            && grep -q '"OPTIONAL_API"' include/asterisk/buildopts.h
+
+        FROM andrius/asterisk:22.8-cert4_debian-trixie@sha256:f3bd01300f0c36c8a7ea5bd9472c4b906d0f85410c3ea429166a0fedac6ad3ee
+        COPY --from=modules /usr/src/asterisk/res/res_audiosocket.so \
+            /usr/src/asterisk/channels/chan_audiosocket.so \
+            /usr/src/asterisk/apps/app_audiosocket.so /usr/lib/asterisk/modules/
+    image: ${RTPBENCH_REGISTRY:?}/rtpbench-ava-asterisk:22.8-cert4${RTPBENCH_BUILD_ASTERISK:-}
+    network_mode: service:target
+    depends_on: [target]
+    volumes:
+      - ./targets/ava/pjsip.conf:/etc/asterisk/pjsip.conf:ro
+      - ./targets/ava/extensions.conf:/etc/asterisk/extensions.conf:ro
+      - ./targets/ava/ari.conf:/etc/asterisk/ari.conf:ro
+      - ./targets/ava/http.conf:/etc/asterisk/http.conf:ro
+      - ./targets/ava/websocket_client.conf:/etc/asterisk/websocket_client.conf:ro
+    healthcheck:
+      test:
+        - CMD-SHELL
+        - >-
+          asterisk -rx 'ari show status' | grep -q Enabled &&
+          asterisk -rx 'module show like chan_audiosocket' | grep -q Running
+      interval: 1s
+      retries: 30
+
+  engine:
+    build:
+      context: https://github.com/hkjarral/AVA-AI-Voice-Agent-for-Asterisk.git#ba52df92bc1cad2c1f277034e1423b80c8f50be4
+      # upstream's Dockerfile, with the base image and every installed package pinned
+      # (upstream pins some of its direct requirements)
+      additional_contexts: { pins: ./targets/ava }
+      dockerfile_inline: |
+        FROM python:3.11-slim-bookworm@sha256:0a310eeecf4e1f5a0743f9a6520c90c88d089c903ca5fd283f501e3a805f5f89 AS builder
+        WORKDIR /usr/src/app
+        RUN apt-get -o Acquire::Retries=5 update && apt-get install -y --no-install-recommends \
+            build-essential && rm -rf /var/lib/apt/lists/*
+        RUN python -m venv /opt/venv
+        ENV PATH="/opt/venv/bin:$$PATH"
+        COPY requirements.txt .
+        COPY --from=pins constraints.txt /tmp/constraints.txt
+        RUN pip install --no-cache-dir -r requirements.txt -c /tmp/constraints.txt
+
+        FROM python:3.11-slim-bookworm@sha256:0a310eeecf4e1f5a0743f9a6520c90c88d089c903ca5fd283f501e3a805f5f89
+        ENV PYTHONDONTWRITEBYTECODE=1 PYTHONUNBUFFERED=1
+        RUN apt-get -o Acquire::Retries=5 update \
+            && DEBIAN_FRONTEND=noninteractive apt-get install -y --no-install-recommends tzdata \
+            && rm -rf /var/lib/apt/lists/*
+        WORKDIR /app
+        RUN groupadd -g 995 asterisk || true && useradd --create-home appuser \
+            && usermod -aG 995 appuser
+        COPY --from=builder /opt/venv /opt/venv
+        COPY --chown=appuser:appuser src/ ./src
+        COPY --chown=appuser:appuser config/ ./config
+        COPY --chown=appuser:appuser main.py ./
+        RUN mkdir -p /app/logs /app/data && chown appuser:appuser /app/logs /app/data
+        ENV PATH="/opt/venv/bin:$$PATH"
+        USER appuser
+        CMD ["python", "main.py"]
+    image: ${RTPBENCH_REGISTRY:?}/rtpbench-ava:ba52df9${RTPBENCH_BUILD_ENGINE:-}
+    network_mode: service:target
+    depends_on:
+      asterisk: { condition: service_healthy }
+    volumes: ["./targets/ava/ai-agent.yaml:/app/config/ai-agent.yaml:ro"]
+    environment:
+      ASTERISK_HOST: 127.0.0.1
+      ASTERISK_ARI_PORT: "8088"
+      ASTERISK_ARI_USERNAME: asterisk
+      ASTERISK_ARI_PASSWORD: asterisk
+      OPENAI_API_KEY: ${OPENAI_API_KEY:?set OPENAI_API_KEY}
+      OPENAI_BASE_URL: ${OPENAI_BASE_URL:-wss://api.openai.com/v1/realtime}
+      OPENAI_MODEL: ${OPENAI_MODEL:-gpt-realtime-2.1}
+      # the golden config's egress pacer: on as shipped, false to measure what it costs
+      AVA_EGRESS_PACER: ${AVA_EGRESS_PACER:-true}
+      # how Asterisk hands AVA the call audio: AudioSocket (TCP) as shipped, externalmedia (RTP)
+      # or websocket (chan_websocket)
+      AVA_TRANSPORT: ${AVA_TRANSPORT:-audiosocket}
+      # websocket's listener password, as in websocket_client.conf
+      ASTERISK_MEDIA_WS_PASSWORD: asterisk
+      LOG_LEVEL: debug
+    # ready once the engine serves its health endpoint
+    healthcheck:
+      test: python -c 'import urllib.request as u; u.urlopen("http://127.0.0.1:15000/health")'
+      interval: 1s
+      retries: 60
+
+networks:
+  egress:

rtpbench/targets/ava/constraints.txt

@@ -0,0 +1,75 @@
+aiohappyeyeballs==2.7.1
+aiohttp==3.14.3
+aiosignal==1.4.0
+annotated-types==0.8.0
+anyio==4.15.1
+ari==0.1.3
+attrs==26.1.0
+azure-cognitiveservices-speech==1.52.0
+azure-core==1.41.0
+certifi==2026.10.18
+cffi==2.1.1
+charset-normalizer==3.5.2
+click==8.5.0
+cryptography==50.0.2
+dnspython==2.8.0
+frozenlist==1.8.0
+google-api-core==2.42.0
+google-api-python-client==2.201.0
+google-auth==2.62.0
+google-auth-httplib2==0.4.4
+googleapis-common-protos==1.75.5
+h11==0.16.0
+httpcore2==2.14.0
+httplib2==0.32.0
+httpx2==2.14.0
+idna==3.20
+Jinja2==3.1.6
+jiter==0.17.0
+jsonschema==4.26.0
+jsonschema-specifications==2025.9.1
+MarkupSafe==3.0.4
+mcp==2.3.0
+mcp-types==2.3.0
+msal==1.39.0
+msgpack==1.2.1
+multidict==6.9.1
+numpy==2.4.6
+openai==3.35.1
+opentelemetry-api==1.45.1
+portalocker==4.4.1
+prometheus_client==0.24.1
+propcache==0.5.4
+proto-plus==1.29.0
+protobuf==7.36.2
+pyasn1==0.6.4
+pyasn1_modules==0.4.2
+pycparser==3.11
+pydantic==2.14.1
+pydantic-settings==2.14.2
+pydantic_core==2.50.1
+PyJWT==2.15.1
+pyparsing==3.3.3
+python-dotenv==1.2.4
+python-multipart==0.0.32
+PyYAML==6.0.3
+referencing==0.37.0
+requests==2.34.2
+resend==2.50.0
+rpds-py==2026.9.1
+sniffio==1.3.1
+sse-starlette==3.5.0
+starlette==1.9.0
+structlog==26.1.0
+swaggerpy==0.2.1
+tenacity==8.2.3
+truststore==0.10.5
+typing-inspection==0.4.4
+typing_extensions==4.16.0
+uritemplate==4.2.0
+urllib3==2.8.0
+uvicorn==0.54.0
+webrtcvad==2.0.10
+websocket-client==1.9.2
+websockets==16.1.1
+yarl==1.25.1

rtpbench/targets/ava/extensions.conf

@@ -0,0 +1,10 @@
+; the caller dials `agent`; AVA's ARI app answers and runs the call
+[from-caller]
+exten => agent,1,Stasis(asterisk-ai-voice-agent)
+ same => n,Hangup()
+
+; AVA's blind_transfer continues the answered call here: Transfer() sends a REFER, waits for
+; its outcome and leaves the call up, so hang up after it either way (AVA's session has ended)
+[transfer]
+exten => operator,1,Transfer(PJSIP/sip:operator@operator)
+ same => n,Hangup()