About this project

AI Native Call Center is an open-source contact center platform where the AI is the default answer rather than an add-on. Inbound calls arrive at FreeSWITCH and are answered by a speech-to-speech voice model that the application terminates itself, with G.711 audio passed through byte-for-byte where the provider accepts it. A flow engine steers the conversation through phases and tools, deciding when a human agent is needed; transfers then route the caller into mod_callcenter queues with full context attached, so the agent's screen pops before the phone rings. The entire system ships as one Go binary containing the REST API, server-sent event stream, and an embedded React web interface, backed by PostgreSQL and FreeSWITCH. FreeSWITCH reads its directory, dialplan, and queues from the database through Lua, so adding extensions, queues, or numbers is a database change rather than a config edit. Every live call is modeled as an actor with a single goroutine as its sole mutator, eliminating shared-state locks. Key capabilities include: bilingual English and Chinese support throughout the interface and bot; a single provider per deployment chosen at startup (Qwen for mainland China, OpenAI elsewhere); versioned personas, rules, and bot voices published together; built-in tools that can refuse gracefully (e.g., queue closed) as conversational responses rather than errors; a unified call record per conversation with one CDR, transcript, and recording regardless of whether the bot, queue, or agent handled it; recordings to filesystem or any S3-compatible store; an agent desktop with presence, softphone bar, and callbacks; and a supervisor wallboard with live queue view and roster. The HTTP API is spec-first with docs/openapi.json as the single source of truth, generating both the Go server and TypeScript client. The project includes seeded data (a team, two queues, six bilingual flows, eighteen simulated customer phones, and a week of history) so the wallboard is populated and a softphone can ring the bot immediately after deployment. Extension points cover new voice providers as profiles, new flows as validated JSON documents, new screens following a binding design system, and external integrations as services behind the flow's backend URL. The project explicitly excludes any cascaded ASR+LLM+TTS pipeline inside the process, reserving that composition for a separate service. Licensed under Apache-2.0.