The open-source switchboard for LLM voice calls

What it does Bolna is an orchestration framework that chains together telephony services, speech-to-text engines, LLMs, and text-to-speech providers into real-time conversational voice agents. It handles the async streaming over websockets so a user on a phone call can talk to an LLM and hear synthesized responses without you writing the handoff logic. You define the pipeline in Python or JSON, and the framework manages the provider coordination.

The interesting bit The project is refreshingly candid about its own maintenance, admitting it has “fluctuated” between open-source and hosted priorities due to time crunches and is actively seeking maintainers. Architecturally, it treats a conversation as an async generator pipeline — transcriber → LLM → synthesizer — and even offers a text-only mode that strips out the audio layers entirely.

Key highlights

• Swappable provider ecosystem: mix and match telephony (Twilio, Plivo), ASR (Deepgram, Azure), LLMs (OpenAI, DeepSeek, Llama via LiteLLM), and TTS (ElevenLabs, Cartesia, AWS Polly, etc.).
• Real-time websocket streaming rather than batch request/response.
• Text-only pipeline mode for chat-style agents without audio hardware.
• Open-source repo serves as the backbone for a closed-source hosted API and no-code dashboard, meaning features land here first.
• Programmatic agent construction in Python using an Assistant class with task-based pipelines.

Caveats

• The README explicitly states the project has suffered maintenance fluctuations and is “actively looking for maintainers,” signaling uncertain long-term support.
• Several telephony providers (Exotel, Vonage) are listed as “coming soon,” so the provider matrix is incomplete.
• Local deployment requires multiple services (telephony server, Bolna server, Redis, ngrok), suggesting operational complexity.

Verdict Use it if you need to mix-and-match voice providers and can tolerate a multi-container local setup that includes Redis and an ngrok tunnel. Avoid it if you want a finished SaaS with a full-time maintenance team.