AI-Powered Call Center Platform

Part of the worlds-biggest-software-project initiative.

An open-source, AI-native voice platform that handles inbound and outbound calls with natural-language understanding, intelligent escalation to human agents, and full-stack contact centre operations -- replacing rigid IVR trees and expensive per-seat CCaaS contracts.

Traditional call centres are caught between two bad options: overstaffing to meet peak demand or deflecting customers into frustrating IVR menu trees. AI-powered voice agents resolve this by handling the full range of routine calls in natural speech, escalating only genuinely complex interactions to human agents with complete context transferred. This project aims to be the first credible open-source alternative in a market dominated entirely by proprietary SaaS.

Why AI-Powered Call Center Platform?

No open-source option exists. Every leading platform -- Retell AI, Bland AI, Replicant, NICE CXone, Five9, Talkdesk, Genesys, Amazon Connect, Twilio Flex, Salesforce Agentforce -- is proprietary SaaS with vendor lock-in. There is no open-source AI voice agent framework of any significance.
Enterprise pricing shuts out smaller teams. Full-stack platforms (NICE, Five9, Talkdesk, Genesys) are enterprise-only with opaque pricing. SMB-accessible tools like CloudTalk charge separately for AI features (AI Voice Agent add-on starts at EUR 350+/month) and lack real-time copilot, QA, and outbound campaign management in one product.
Per-minute costs compound at scale. Retell AI charges $0.07+/min, Bland AI $0.09-0.14/min after add-ons, Twilio Flex AI Copilot $0.045/min voice. For contact centres handling millions of minutes monthly, these costs become a significant operational expense that a self-hosted solution could substantially reduce.
Data residency is unsolved. Organisations in financial services, government, and healthcare with strict data residency requirements cannot self-host any leading call centre AI platform. A self-hostable deployment option addresses a real compliance gap.
AI decisions are opaque. Most platforms provide limited explainability for why the AI escalated, routed, or responded as it did. Transparent, auditable AI decision logs are an underserved requirement across the industry.

Key Features

Natural-Language Voice AI Agent

LLM-powered conversational AI handling open-ended customer speech with multi-turn dialogue
Sub-800 ms end-to-end latency pipeline: streaming STT + LLM inference with prompt caching + streaming TTS
Intent classification and entity extraction (account numbers, order IDs, etc.)
Knowledge base retrieval-augmented responses with hallucination constraints
Knowledge gap detection: surfaces unanswered questions to improve the knowledge base over time

Intelligent Escalation and Agent Assist

Detect frustration, complexity, or explicit requests for a human agent
Transfer calls with full transcript, extracted entities, and recommended resolution path
Real-time agent assist copilot panel with suggested responses, knowledge base articles, and compliance prompts
Auto-fill CRM fields from call content during human-handled calls

Outbound Campaign Management

Batch scheduling for appointment reminders, payment collection, satisfaction surveys, and proactive notifications
Real-time DNC list checking before each call attempt
TCPA and GDPR consent verification and immutable consent logging
Configurable call-time windows per geographic region

Speech Analytics and Quality Assurance

100% automated QA scoring across all calls (sentiment, compliance adherence, resolution quality)
Real-time call transcription with speaker diarisation
Automated post-call summaries with CRM auto-update
Coaching workflows triggered by QA scores
Supervisor alerts on sentiment-triggered escalation risk

Dashboards, Compliance, and Integration

Live supervisor dashboard: queue depth, agent utilisation, AI containment rate, AHT, CSAT
PCI DSS pause-resume recording for card data; GDPR-aligned consent capture at call start
Configurable call retention and deletion policies
Bi-directional CRM connectors for Salesforce, HubSpot, Zendesk, and ServiceNow
REST API and webhooks for custom integrations

AI-Native Advantage

Unlike traditional IVR systems that force callers through rigid menu trees, this platform deploys LLM-based voice agents that understand natural language, maintain multi-turn conversational context, and retrieve answers from a connected knowledge base rather than hallucinating. AI augmentation extends beyond the caller-facing agent: 100% of calls are scored automatically for quality and compliance (replacing the industry-standard 2-5% manual sampling), after-call work is eliminated through automated summarisation and CRM field population, and knowledge gaps detected during calls feed directly back into content improvement. Multi-LLM routing -- directing different sub-tasks (intent detection, knowledge retrieval, response generation) to the most cost-effective model -- is a first-class capability not exposed by any incumbent.

Tech Stack and Deployment

Voice pipeline: Streaming speech-to-text (Deepgram, AssemblyAI, or self-hosted Whisper on GPU), low-latency LLM inference with prompt caching for conversation context, and streaming text-to-speech (Cartesia, ElevenLabs Flash, or Play.ht) that begins audio generation before the full response text is ready.

Telephony: SIP trunking via Twilio Voice, Vonage, or Bandwidth for PSTN connectivity. WebRTC for browser-based agent desktops. Media servers (Asterisk, FreeSWITCH) for call mixing during conference escalations.

Session state: Externalised to Redis or DynamoDB so conversation context survives process restarts mid-call.

Standards: Built on open IETF standards -- SIP (RFC 3261), RTP (RFC 3550), STIR/SHAKEN (RFC 8226/8588), WebRTC (RFC 8825) -- and W3C VoiceXML. No proprietary protocol dependencies.

Deployment modes: Self-hosted (Docker/Kubernetes), cloud, or hybrid. Self-hostable compliance tier for organisations with strict data residency requirements (financial services, government, healthcare).

Market Context

The AI call centre software market is crowded and mature by 2026, with platforms ranging from pure voice AI stacks (Retell AI, Bland AI, Replicant) to AI-augmented CCaaS providers (NICE, Five9, Talkdesk, Genesys, Amazon Connect, Twilio Flex) and CRM-native solutions (Salesforce Agentforce, Microsoft Dynamics 365 Contact Center). Contact centres deploying AI voice agents report 40-70% containment rates on high-volume call types. The market is driven by labour cost pressure (agent turnover averages 30-45% annually), 24/7 availability requirements, and the availability of low-latency, high-accuracy speech models. All leading platforms are proprietary SaaS with enterprise pricing, leaving a clear gap for an open-source, self-hostable alternative.

Project Status

This project is in the research and specification phase. Contributions, feedback, and domain expertise are welcome.

Contributing

We welcome contributions from developers, domain experts, and potential users. See CONTRIBUTING.md for guidelines.

Important: All contributions must be your own original work or clearly attributed open-source material with a compatible licence. Copyright infringement and licence violations will not be tolerated and will result in immediate removal of the offending contribution. If you are unsure whether a piece of code, text, or other material is safe to contribute, open an issue and ask before submitting.

Licence

Licence to be determined. See discussion for context.