# Pinecall

> Pinecall is an open-source framework for building and running voice applications, not just agents. It handles call routing, testing, supervision, and failovers, and is designed for developers and teams deploying voice apps to production.

- Type: AI tool
- Page: https://completeaitraining.com/ai-tools/pinecall/
- Published: 2026-11-16
- Updated: 2026-11-18
- Author: Jeroen Erne
- Jobs: IT and development, Operations
- Topics: Open Source, Developer Tools, Coding
- Reading time: 2 min

### About Pinecall

Pinecall is an open-source platform for building and running voice AI agents, released under the Apache-2.0 license. It's built on top of LiveKit for real-time communication and provides a framework for creating voice apps using TypeScript, Python, or Ruby, a form-based Studio, or via MCP from a coding assistant. The platform includes tools for testing with simulated callers, logging and judging every call, and keeping a human supervisor one tap away from the conversation.

### Review

Pinecall tackles the infrastructure around voice agents rather than the speech-to-text or model integrations themselves, which it delegates to LiveKit. The platform is new this week and has run in production for a handful of companies, so it comes with the rough edges of early software. That said, the focus on testing, observability, and failure handling addresses real gaps that teams encounter when voice agents interact with actual callers.

#### Key Features

- **Multi-surface agent building.** Agents can be written in TypeScript, Python, or Ruby, configured through a form-based Studio, or generated by a coding assistant over MCP. All paths run on the same runtime.
- **Simulated caller testing.** Written conversation tests and simulated callers that speak over a real audio line, including noise and packet loss, let teams test agent behavior before deploying to real users.
- **Call logging and judging.** Every call receives a verdict at hang-up and a detailed log entry by entry. A broken call can be converted into a test case for regression testing.
- **Supervisor intervention.** A human supervisor can listen in on a live call, whisper to the agent, or take over the line entirely from the console or a phone, then hand it back.
- **Self-hosted runtime with open models.** A single command spins up the full runtime locally using open models-Whisper large-v3-turbo, Gemma 4 or NVIDIA Nemotron on Ollama, and Kokoro for voice-on NVIDIA GPUs or Apple silicon, with no external vendor accounts required.

#### Pricing and Value

Pinecall is open source under Apache-2.0. The sandbox environment on pinecall.io includes 120 free minutes with no credit card required. The platform can be deployed on your own Kubernetes cluster using provided Terraform configurations and Helm charts, or run on Pinecall's own cloud infrastructure. Pricing for a managed cloud plan is not yet defined.

#### Pros

- Open-source codebase under Apache-2.0 means no vendor lock-in for the core platform.
- Simulated callers that interrupt and talk fast create more realistic test conditions than scripted playback.
- Worker failover transfers an in-progress call to another worker with full conversation history if the original worker dies mid-call.
- Deployments drain live calls gracefully instead of cutting them, and speech-to-text plus voice synthesis fall back to alternate vendors on failure.
- Runs entirely offline with open models, which matters for teams with data locality or cost constraints.

#### Cons

- Production track record is limited to a handful of companies, so long-term stability patterns are not yet established.
- The platform is not well suited for teams that want a fully managed, hands-off voice agent service with no infrastructure responsibility.
- As a newly launched project, documentation and community support are likely thin compared to more mature alternatives.

Pinecall fits teams that already have engineering capacity and want control over their voice agent infrastructure, especially those who need rigorous testing and observability before putting agents in front of real callers. The local-only runtime with open models also makes it relevant for prototyping without vendor costs. Organizations looking for a battle-tested, managed solution with extensive production history should wait for the platform to mature.

---

This is the Markdown form of https://completeaitraining.com/ai-tools/pinecall/. The whole site is mapped at https://completeaitraining.com/llms.txt.
