Vibeleaderboard
← Back to Tools
Index / agent

Call Center AI

https://github.com/microsoft/call-center-ai
Visit github.com
Category
AI Agents
Rank
No. 1014Tools index
Pricing
Open Source
Type
AGENT
Builder
microsoft
GitHub
6.6k stars
Latest release
v17.4.2
Date

About

Microsoft's reference architecture for placing AI-powered phone calls — your bot dials out (or callers dial in) and an LLM drives the conversation.

What it does

It coordinates live speech, language handling, model responses, SMS, and human transfer around a configurable call workflow. Each conversation can populate typed claim fields, reminders, next actions, summaries, and a browsable report.

Why it's ranked here

This is a substantial reference implementation, not a thin telephony demo. It connects real-time audio, structured data capture, retrieval, persistence, monitoring, and agent fallback. The explicit proof-of-concept status keeps it from being a production-ready choice.

What's good

The design preserves useful business output beyond a transcript: validated claim data, reminders, summaries, satisfaction signals, and recommended next actions. Readiness checks cover cache, storage, search, and SMS. Calls can resume after disconnections, while jittered retries improve Twilio HTTP reliability.

Tradeoffs

Deployment brings a large Azure footprint, including communication, speech, translation, search, queues, storage, Cosmos DB, Redis, Event Grid, monitoring, and model resources. The project requires Python 3.13 and several command-line tools. Its own documentation says it is not intended for production.

How to use it well

Use it as an engineering starting point for testing structured, low-to-medium-complexity phone workflows such as claims intake or support triage. Start in Codespaces, customize prompts and claim fields, then validate latency, escalation, privacy, and failure recovery. It does not replace production hardening or operational ownership.

Technical notes+

app/main.py exposes a FastAPI service with REST call endpoints, HTML reports, health probes, WebSocket media handling, Azure Communication Services callbacks, and queue consumers. app/models/call.py validates dynamic claim data and stores messages, reminders, synthesis, and next actions. app/helpers/http.py shares an asynchronous HTTP session and adds jittered Twilio retries. pyproject.toml requires Python 3.13, declares Azure, FastAPI, Redis, Twilio, telemetry, and testing dependencies, and supports uv installation. Makefile provides static checks, Pytest execution, multi-architecture container builds, Bicep deployment, logs, tunnels, and confirmed teardown.

Observed

License
Apache-2.0
Primary language
Python
Runtime
Python 3.13 or newer
Interfaces
REST API, WebSocket media stream, HTML reports, telephony callbacks, and SMS webhook
Packaging and install
Python project installed with uv; pre-built container image and GitHub Codespaces quick start are available
Platform support
Azure deployment using Container Apps, Communication Services, Cognitive Services, Azure OpenAI, Cosmos DB, Event Grid, Storage, AI Search, Redis, and Application Insights
Verification surface
Ruff, Pyright, Bicep linting, and Pytest are wired into Make targets

Read from README.md, Makefile, pyproject.toml, app/main.py, app/models/call.py, app/models/next.py, app/helpers/http.py, app/models/claim.py, app/models/error.py, app/helpers/cache.py, app/helpers/config.py, app/models/message.py, app/helpers/logging.py, app/models/reminder.py.

What it can do

  • Make outbound AI-powered phone calls

    Phone number and conversation parameters → Automated phone conversation

  • Receive and handle inbound phone calls

    Incoming phone call → AI-driven conversation response

  • Generate conversational responses using LLM

    Caller's speech and conversation context → Natural language responses

  • Convert speech to text

    Caller's voice audio → Text transcription

  • Convert text to speech

    AI-generated text responses → Synthesized voice audio

  • Manage call flow and conversation state

    Conversation history and current context → Updated conversation state and next actions

Tags

voice-agentphonemicrosofttelephonyai-agent

Tech Stack

Python

Comments (0)

No comments yet

Editorially curated, with community endorsements as a secondary signal. Corrections welcome.