v0.4.0

What's Changed

• vk test --target dfcx|ces: run tests against a live Google agent by @C4rohan in #41
• Fix the October review findings (paths, transfers, outbound calls, keys in URLs) by @C4rohan in #43
• Test a phone agent for free from your laptop by @C4rohan in #42
• Rebuild the UI after merging #42 and #43 by @C4rohan in #44
• Evals tab: real phone calls through Google Voice, played by a Test Caller and graded by @C4rohan in #45
• Fix the UI build check on dev, and stop it from breaking again by @C4rohan in #46
• Import test scripts as eval test cases; clearer Start eval; v0.3.0 by @C4rohan in #47
• Dev by @C4rohan in #48
• Enhance testing framework for Google agents and improve UI by @C4rohan in #49
• Hotfix: timing-sensitive realtime test failing on prod by @C4rohan in #50
• Grade real calls automatically; p95 response time and cost per call by @C4rohan in #51
• A tougher Test Caller: noise, voices, cut-ins, pauses, caller-side timing, replay by @C4rohan in #52
• Gemini Live speech-to-speech next to OpenAI Realtime by @C4rohan in #53
• Fix a race that could end a speech-to-speech call before its goodbye by @C4rohan in #54
• README: a 30-second demo, free phone evals up front, and a no-keys start by @C4rohan in #55
• Fix the real cause of the early hang-up in speech-to-speech calls by @C4rohan in #56
• Automate call grading, enhance testing, and improve README demo by @C4rohan in #57
• Merge prod into qa (unblocks #58) by @C4rohan in #59
• Two fixes from trying --free: model name and local speech downloads by @C4rohan in #60
• Simpler shipping: one script, and CI that cleans up after itself by @C4rohan in #61
• Fix model handling and improve local speech model downloads by @C4rohan in #62
• Automate call grading, enhance testing, and improve README demo by @C4rohan in #58

Full Changelog: v0.3.0...v0.4.0

Lampline v0.3.0

Lampline v0.3.0: test real phone calls, graded
Lampline is an open-source, self-hosted platform for building voice agents, like a CX Agent Studio you run yourself. It's the first release since v0.1.0, so it also includes everything listed under v0.2.0 in the changelog: free local speech that answers faster (Moonshine, Parakeet), transfers to a person with the call's details as SIP headers, users and roles, call history, and speech-to-speech. This release is about testing: real phone calls to your agent, played by an AI caller and graded.

Test real calls
Evals tab. A Test Caller phones your voice agent, plays a persona with a goal, and the call is graded against your pass criteria, with a transcript and a reason for each check.
Google Voice (free): the new Lampline Phone Bridge extension for Chrome and Edge connects the call's audio to the Test Caller inside the browser. No speakers or cables, on Windows or Mac. You press Call yourself.
Phone on speaker: use any phone, or a virtual audio cable, with the microphone and speaker you choose.
A call where nothing was heard shows "Didn't run", so a bad connection never counts as a failed agent.
Import test scripts. Paste or upload Hammer, Cyara, spreadsheet or plain-text tests, and your model turns each one into a test case you can run. Keypad steps are said as digits.
Test Caller template. Start steps can wait for the other side to speak first.
Test Google agents
vk test --target dfcx|ces runs your test files against a live Dialogflow CX agent or CX Agent Studio app, or side by side with Lampline (--target engine,dfcx).
vk test can let a model play the caller.
Security and reliability
Saving an agent can't write outside its folder. A remote save can't change where calls are transferred.
Outbound calls never ring twice: a contact is claimed before the call, a timeout answers 504, and the API takes an Idempotency-Key.
The API key never goes in a URL. "Only on this computer" fails closed behind proxies.
Carriers' redelivered webhooks still connect transfers. WhatsApp messages are answered once.
Live conversations are visible only to the key that started them. The public health check shows no setup details.
Upgrade

git pull (or pull the new image), then restart the engine. Your eval history upgrades in place. To use Google Voice, load the extension folder in chrome://extensions (Developer mode → Load unpacked).

Full list: CHANGELOG.md · How to run evals: docs/EVALS.md

Lampline v0.1.0

The first release of Lampline: a CX Agent Studio–style voice agent platform you run yourself.

Highlights

• Import from Dialogflow CX and CX Agent Studio: flows, pages, forms, routes, playbooks and tools; and export back to both.
• The engine on Pipecat: a drag-and-drop builder where LLM steps and scripted pages mix in one agent, text chat, browser calls and phone calls.
• 20 model providers: Claude, OpenAI-compatible, Gemini, Vertex AI, Azure OpenAI, Bedrock, Groq, Mistral, DeepSeek, Grok, Together, Fireworks, OpenRouter, Cerebras, NVIDIA NIM, Qwen, SambaNova, Nebius, Novita and Ollama.
• Phone and SIP: Twilio, Telnyx, and any SIP trunk, PBX or SBC through jambonz.
• Data and knowledge of its own: record collections and local document search.
• Tools and triggers: HTTP, MCP and connector tools; schedules, Telegram, WhatsApp and email.
• vk: lint with a personal-data scan, conversation tests, starter templates and coding-agent skills.
• One Docker container for local and on-prem, non-root and key-protected.

Get started

git clone https://github.com/C4rohan/lampline && cd lampline
scripts/docker-up.sh

Coming soon: Google Cloud deployment, and Genesys, Avaya, Cisco and other contact centers (roadmap).