Skip to content
Back to search

lampline

MarketplaceWorks with Claude CodeGitHub

Build, test, fix and review phone agents with Lampline's vk CLI

Safety scan

A static text scan, not a guarantee. Select the badge to see what was scanned and every finding.

Checked today
Badge for your README
findsafeskills safety rating

Markdown

[![findsafeskills safety rating](/api/listings/c4rohan--lampline/badge.svg)](/listings/c4rohan--lampline)

HTML

<a href="/listings/c4rohan--lampline"><img src="/api/listings/c4rohan--lampline/badge.svg" alt="findsafeskills safety rating" /></a>

Install

Install instructions from the README

As written by the repo's author. findsafeskills has not verified these steps.

**Just the tests** (Python 3.10+; no Docker, no clone):

pipx install lampline                                     # the vk command
vk templates install front-desk --dir agents/front-desk   # a starter agent with tests
vk lint agents/front-desk                                 # offline checks

From here you can [test your Dialogflow CX agent in 10 minutes](docs/TEST-DIALOGFLOW.md), or start the engine below and run `vk test agents/front-desk`.

**The whole platform** (builder, test calls, phone calls). Needs Docker with Compose v2 (Docker Desktop is fine). **No API keys or accounts:**

git clone https://github.com/C4rohan/lampline && cd lampline
scripts/docker-up.sh --free          # everything on this computer: Ollama, Moonshine and Kokoro

`--free` runs the model (Qwen3 4B in Ollama) and speech (Moonshine listening, Kokoro speaking) on your own computer. The first start downloads about 3 GB and shows progress; after that it starts in seconds. Replies are slower and simpler than a cloud model's, and you can switch to any provider on the Integrations page whenever you like. On a Mac, the native route is faster, because Ollama uses the GPU outside Docker: install [Ollama](https://ollama.com), then `scripts/engine-up.sh --free`.

Have API keys? Run `scripts/docker-up.sh` without `--free`. Either way it prints a link like `http://localhost:7860/#/?key=…`: open it once and the page remembers the key. Pick your model and voice on the Integrations page (or put keys like `ANTHROPIC_API_KEY`, `GEMINI_API_KEY`, `DEEPGRAM_API_KEY` in `.env`; see `.env.example`). Your agents are in `./agents` (edit them, keep them in git) and everything else the engine keeps, settings with your keys, records, knowledge and downloaded models, is in `./.vpe-data`. `scripts/docker-up.sh --down` stops it; `--logs` follows its log. Plain `docker compose up -d --build` works too once `.env` has `VPE_DOCKER_API_KEY`. Details: [engine/README.md](engine/README.md#docker).

Prefer no Docker? `scripts/engine-up.sh` runs the engine natively (Python 3.11+) at http://localhost:7860, which also lets browser calls reach it directly. Free, private speech: Integrations → Whisper and Kokoro → **Download models**, or Moonshine for listening, which transcribes while the caller talks.

What was scanned

The files the safety scan read, as written by the repository's author and shown as plain text. Lines with a finding are marked.

.claude-plugin/marketplace.json · Manifest · 29 lines
  1. {
  2. "name": "lampline",
  3. "description": "Build, test, fix and review phone agents with Lampline's vk CLI",
  4. "owner": {
  5. "name": "Rohan Salvi"
  6. },
  7. "plugins": [
  8. {
  9. "name": "lampline",
  10. "source": "./",
  11. "description": "Skills to build, test, fix and review phone agents with vk, including testing existing Dialogflow CX and CX Agent Studio agents",
  12. "homepage": "https://github.com/C4rohan/Lampline",
  13. "license": "Apache-2.0",
  14. "keywords": [
  15. "voice-agents",
  16. "phone",
  17. "testing",
  18. "dialogflow-cx",
  19. "pipecat"
  20. ],
  21. "skills": [
  22. "./cli/src/vk/skills/voice-agent-builder",
  23. "./cli/src/vk/skills/voice-agent-reviewer",
  24. "./cli/src/vk/skills/voice-agent-fixer",
  25. "./cli/src/vk/skills/dialogflow-tester"
  26. ]
  27. }
  28. ]
  29. }
README.md · README · 455 lines
  1. <p align="center">
  2. <picture>
  3. <source media="(prefers-color-scheme: dark)" srcset="docs/brand/lampline-logo-dark.png">
  4. <img src="docs/brand/lampline-logo-light.png" alt="Lampline" width="280">
  5. </picture>
  6. </p>
  7. <h3 align="center">Test and run phone agents on your own machine.</h3>
  8. <p align="center">
  9. Write conversation tests for your phone agents, including <b>Dialogflow CX</b> and <b>CX Agent Studio</b> agents,<br>
  10. run them from <b>Claude Code</b> or CI, and have real phone calls graded by an AI caller.<br>
  11. Open source, and free to run with no API keys.
  12. </p>
  13. <p align="center">
  14. <a href="https://github.com/C4rohan/lampline/actions/workflows/ci.yml"><img src="https://github.com/C4rohan/lampline/actions/workflows/ci.yml/badge.svg?branch=dev" alt="CI"></a>
  15. <a href="LICENSE"><img src="https://img.shields.io/badge/license-Apache--2.0-blue" alt="Apache-2.0"></a>
  16. <img src="https://img.shields.io/badge/tests-Dialogflow_CX_%C2%B7_CX_Agent_Studio-2f6fb0" alt="Tests Dialogflow CX and CX Agent Studio agents">
  17. <img src="https://img.shields.io/badge/Claude_Code-plugin-d97757" alt="Claude Code plugin">
  18. <img src="https://img.shields.io/badge/models-20_providers-f2b84b" alt="20 model providers">
  19. <img src="https://img.shields.io/badge/runs-local_%26_on--prem-15212b" alt="Runs local and on-prem">
  20. <a href="https://codespaces.new/C4rohan/lampline"><img src="https://github.com/codespaces/badge.svg" alt="Open in GitHub Codespaces" height="20"></a>
  21. </p>
  22. <p align="center">
  23. <a href="#-quick-start"><b>🚀 Quick start</b></a> &nbsp;·&nbsp;
  24. <a href="#-use-it-from-claude-code-cursor-or-codex">🤖 From Claude Code</a> &nbsp;·&nbsp;
  25. <a href="docs/TEST-DIALOGFLOW.md">🧪 Test a Dialogflow agent</a> &nbsp;·&nbsp;
  26. <a href="#-test-any-phone-agent-for-free">📋 Free phone evals</a> &nbsp;·&nbsp;
  27. <a href="#-bring-your-dialogflow-cx-and-cx-agent-studio-agents">🔁 Import from Google</a> &nbsp;·&nbsp;
  28. <a href="#-a-quick-tour">👀 Tour</a> &nbsp;·&nbsp;
  29. <a href="#-models">🧠 Models</a> &nbsp;·&nbsp;
  30. <a href="#-phone-calls-and-sip">📞 Phone and SIP</a> &nbsp;·&nbsp;
  31. <a href="docs/ROADMAP.md">🗺️ Roadmap</a> &nbsp;·&nbsp;
  32. <a href="engine/README.md">📘 Engine docs</a> &nbsp;·&nbsp;
  33. <a href="cli/README.md">🧩 <code>vk</code> CLI</a>
  34. </p>
  35. <br>
  36. <p align="center">
  37. <img src="docs/images/hero.gif" width="900" alt="30 seconds: import a Dialogflow CX export, open it in the builder, chat with it while its steps light up, then a real phone call to it played by the Test Caller and graded Passed">
  38. <br><sub>Import a Dialogflow CX agent, test it, and grade a real phone call to it. Sample agent and made-up data; replies from Lampline's test model.</sub>
  39. </p>
  40. ```bash
  41. pipx install lampline # the vk command: lint, test, import, export
  42. ```
  43. Have a Dialogflow CX agent? [Test it in 10 minutes](docs/TEST-DIALOGFLOW.md).
  44. Want the whole platform? [Quick start](#-quick-start): one command, no API keys.
  45. **Test.** An agent's tests are plain YAML: what the caller says (or a model
  46. playing the caller), and what must happen: which steps the call reaches, what
  47. the agent must or must never say, which tools it calls. `vk test` runs them on
  48. Lampline, or against a live Dialogflow CX or CX Agent Studio agent, and tells
  49. you exactly what went wrong:
  50. ```text
  51. $ vk test agents/front-desk
  52. PASS billing question goes to the billing team (2 turn(s), 0.8s) Greeting → Router → Billing → Transfer to staff
  53. PASS emergency gets the 911 instruction (1 turn(s), 0.7s) Greeting → Router
  54. FAIL caller asks about Saturday hours (1 turn(s), 0.8s) Greeting → Router
  55. - expected to reach 'Hours and location'; steps were Greeting → Router
  56. - expected the agent to say 'noon'
  57. ```
  58. <sub>A starter template run against Lampline's stand-in test model, which is why one test fails.</sub>
  59. **Fix and review from your coding agent.** Lampline ships skills for Claude
  60. Code, Cursor and Codex that build an agent with tests, fix failing tests
  61. without breaking passing ones, test your Dialogflow agents, and review a change
  62. before it goes live ([set up](#-use-it-from-claude-code-cursor-or-codex)).
  63. **Run it.** The same agents take real calls on hardware you control: a
  64. drag-and-drop builder where LLM steps and scripted pages mix, HTTP and MCP
  65. tools, records and document search, and phone calls over Twilio, Telnyx or any
  66. SIP trunk through jambonz. Import from Dialogflow CX and CX Agent Studio, or
  67. export back, whenever you like. Built on [Pipecat](https://github.com/pipecat-ai/pipecat).
  68. <table>
  69. <tr>
  70. <td width="50%" valign="top"><b>🏠 Local first</b><br>One command on a laptop, a GPU box or an on-prem server. No cloud account needed.</td>
  71. <td width="50%" valign="top"><b>🔒 Private by default</b><br>No telemetry, bound to localhost, a key on every request, and keys never in your agents or the repo.</td>
  72. </tr>
  73. <tr>
  74. <td valign="top"><b>🧠 Any model</b><br>Claude, OpenAI, Gemini, Vertex AI, Azure OpenAI, Bedrock, Groq, Mistral, DeepSeek, Grok and more, or Ollama on your own hardware.</td>
  75. <td valign="top"><b>🧩 Agents as code</b><br>Every agent is a folder of YAML and Markdown: reviewed in git, tested with <code>vk</code>, exported back to Google's tools whenever you like.</td>
  76. </tr>
  77. </table>
  78. ## 🚀 Quick start
  79. **Just the tests** (Python 3.10+; no Docker, no clone):
  80. ```bash
  81. pipx install lampline # the vk command
  82. vk templates install front-desk --dir agents/front-desk # a starter agent with tests
  83. vk lint agents/front-desk # offline checks
  84. ```
  85. From here you can [test your Dialogflow CX agent in 10 minutes](docs/TEST-DIALOGFLOW.md),
  86. or start the engine below and run `vk test agents/front-desk`.
  87. **The whole platform** (builder, test calls, phone calls). Needs Docker with
  88. Compose v2 (Docker Desktop is fine). **No API keys or accounts:**
  89. ```bash
  90. git clone https://github.com/C4rohan/lampline && cd lampline
  91. scripts/docker-up.sh --free # everything on this computer: Ollama, Moonshine and Kokoro
  92. ```
  93. `--free` runs the model (Qwen3 4B in Ollama) and speech (Moonshine listening,
  94. Kokoro speaking) on your own computer. The first start downloads about 3 GB and
  95. shows progress; after that it starts in seconds. Replies are slower and simpler
  96. than a cloud model's, and you can switch to any provider on the Integrations
  97. page whenever you like. On a Mac, the native route is faster, because Ollama
  98. uses the GPU outside Docker: install [Ollama](https://ollama.com), then
  99. `scripts/engine-up.sh --free`.
  100. Have API keys? Run `scripts/docker-up.sh` without `--free`. Either way it prints
  101. a link like `http://localhost:7860/#/?key=…`: open it once and the page
  102. remembers the key. Pick your model and voice on the Integrations page (or put keys like `ANTHROPIC_API_KEY`, `GEMINI_API_KEY`, `DEEPGRAM_API_KEY` in
  103. `.env`; see `.env.example`). Your agents are in `./agents` (edit them, keep them
  104. in git) and everything else the engine keeps, settings with your keys, records,
  105. knowledge and downloaded models, is in `./.vpe-data`.
  106. `scripts/docker-up.sh --down` stops it; `--logs` follows its log. Plain
  107. `docker compose up -d --build` works too once `.env` has `VPE_DOCKER_API_KEY`.
  108. Details: [engine/README.md](engine/README.md#docker).
  109. Prefer no Docker? `scripts/engine-up.sh` runs the engine natively (Python 3.11+)
  110. at http://localhost:7860, which also lets browser calls reach it directly.
  111. Free, private speech: Integrations → Whisper and Kokoro → **Download models**,
  112. or Moonshine for listening, which transcribes while the caller talks.
  113. ## 📋 Test any phone agent for free
  114. Lampline's **Evals** tab phones a voice agent over a real phone line, plays the
  115. caller with an AI **Test Caller**, and grades the call against what should
  116. happen. It works on any agent with a phone number: one built in Lampline,
  117. Dialogflow CX, CX Agent Studio, Vapi, Retell, or your IVR.
  118. <table>
  119. <tr>
  120. <td width="52%" valign="top">
  121. <picture>
  122. <source media="(prefers-color-scheme: dark)" srcset="docs/images/evals-dark.webp">
  123. <img src="docs/images/evals-light.webp" alt="A graded eval: the call's transcript, the agent's response time as the caller heard it, and each check passed">
  124. </picture>
  125. </td>
  126. <td valign="top">
  127. - **Free calls through Google Voice.** The Lampline Phone Bridge extension (Chrome or Edge) connects the call's audio to the Test Caller inside the browser: no cables, Windows or Mac. Or use any phone on speaker.
  128. - **Graded, with reasons.** You write who is calling, what they want and what must happen; your model grades the transcript, check by check. A call that heard nothing shows *Didn't run*, never *Failed*.
  129. - **Tough callers.** Background noise, accents, interruptions, long pauses, hesitant or rambling speech.
  130. - **The agent's speed, as callers hear it.** Typical and p95 response time, phone line included.
  131. - **Bring your test scripts.** Paste Hammer, Cyara or spreadsheet tests and they become runnable cases. Replay any failed call.
  132. - **Improve an agent from its failures.** Lampline runs your test cases as text chats, rewrites the prompts of the steps where checks failed, and keeps a change only if more checks pass and nothing that passed breaks. You review the before and after, then apply.
  133. [How to run evals →](docs/EVALS.md)
  134. </td>
  135. </tr>
  136. </table>
  137. Only call agents you own or are allowed to test. You press Call yourself:
  138. Google Voice doesn't allow automated dialing.
  139. ## 🤖 Use it from Claude Code, Cursor or Codex
  140. Lampline agents are plain files (`agent.yaml` and `prompts/*.md`), and `vk`
  141. lints and tests them from the terminal, so a coding agent can build, test and
  142. review a phone agent for you. Four skills teach it how:
  143. - **voice-agent-builder:** asks what the agent should do, writes the tests
  144. first, builds the folder, and runs `vk lint` and `vk test` until they pass
  145. - **voice-agent-fixer:** makes failing tests pass by changing only the
  146. prompts or routes behind them, and keeps a change only if more tests pass
  147. and none breaks
  148. - **dialogflow-tester:** writes tests for an existing Dialogflow CX or CX
  149. Agent Studio agent and runs them against it live, or side by side with its
  150. Lampline copy
  151. - **voice-agent-reviewer:** checks a change before it goes live: lint, the
  152. diff, the tests, and a privacy review, then asks you to approve
  153. **Claude Code**, from the plugin marketplace in this repo:
  154. ```
  155. /plugin marketplace add C4rohan/lampline
  156. /plugin install lampline@lampline
  157. ```
  158. On Claude Code 2.1.275 or later, one line does both:
  159. `/plugin install lampline --marketplace C4rohan/lampline`.
  160. **Cursor, Codex and other agents** read [`AGENTS.md`](AGENTS.md). To copy the
  161. skills into a project instead, run `vk skills install --target claude`
  162. (or `--target agents`, `--target gemini`).
  163. **Claude Desktop, Cursor and other MCP clients:** `vk mcp` serves your agent
  164. folders as tools: list, lint, run tests, read and change a step's prompt, and
  165. see which real calls failed their checks
  166. ([set up](cli/README.md#mcp-server)).
  167. Then ask, for example: *"Build a phone agent for a dental office that books
  168. cleanings, with tests"*, *"Make the failing tests in agents/front-desk pass"*,
  169. *"Write regression tests for my Dialogflow CX agent"*, or *"Review my change
  170. to agents/front-desk before it goes live"*.
  171. ## 🔁 Bring your Dialogflow CX and CX Agent Studio agents
  172. Export an agent from Google's console, drop the .zip on **New agent → Import**, and
  173. it opens in the builder: flows, pages, forms, routes, playbooks and tools carried
  174. over, with a report of anything to check before it takes calls. It works the
  175. other way too: **Export** writes any agent back as a Dialogflow CX package or a
  176. CX Agent Studio app.
  177. <table>
  178. <tr>
  179. <td width="50%" valign="top">
  180. <picture>
  181. <source media="(prefers-color-scheme: dark)" srcset="docs/images/import-done-dark.webp">
  182. <img src="docs/images/import-done-light.webp" alt="Importing a Dialogflow CX export: what came over, and what to review">
  183. </picture>
  184. <p><b>📥 Import.</b> A Dialogflow CX package or a CX Agent Studio app (JSON or YAML). The conversion runs on your computer and never runs code from the export.</p>
  185. </td>
  186. <td width="50%" valign="top">
  187. <picture>
  188. <source media="(prefers-color-scheme: dark)" srcset="docs/images/import-builder-dark.webp">
  189. <img src="docs/images/import-builder-light.webp" alt="The imported agent in the builder: flows, scripted pages and a playbook">
  190. </picture>
  191. <p><b>✏️ Ready to edit.</b> Pages become scripted pages, playbooks become LLM steps, flows group them, and webhooks and OpenAPI tools become tools.</p>
  192. </td>
  193. </tr>
  194. </table>
  195. | From Dialogflow CX | Becomes |
  196. |---|---|
  197. | Flows and pages | Flows of [scripted pages](engine/README.md#scripted-pages-and-flows): entry messages said word for word |
  198. | Form parameters, entity types | Form fields with their types, choices and synonyms, patterns and keypad input |
  199. | Intent and condition routes, no-match handlers | Transfers the model decides, rules, and "no match" transfers |
  200. | Playbooks | LLM steps with their goal, guidelines and steps |
  201. | Webhooks, OpenAPI tools, data store tools | Tools (webhooks answer with a mock until connected) and local knowledge search |
  202. | From CX Agent Studio | Becomes |
  203. |---|---|
  204. | Root agent and child agents | A root agent with sub-agents and transfers both ways |
  205. | Global instruction, variables | Global instructions and context variables |
  206. | OpenAPI toolsets, Python tools, data stores | HTTP tools, tools with example answers, and local knowledge search |
  207. From the terminal: `vk import auto export.zip` and `vk export dfcx|ces agents/<name>`.
  208. The full mapping, both ways, and what doesn't carry over yet:
  209. [cli/README.md](cli/README.md#import).
  210. ## 👀 A quick tour
  211. <p align="center">
  212. <img src="docs/images/demo.gif" alt="A 25-second tour: the agent list, the builder with a transfer to a person, speech options in Integrations, and the Calls page with a call's details" width="880">
  213. <br><sub>Sample data only.</sub>
  214. </p>
  215. <table>
  216. <tr>
  217. <td width="50%" valign="top">
  218. <picture>
  219. <source media="(prefers-color-scheme: dark)" srcset="docs/images/test-dark.webp">
  220. <img src="docs/images/test-light.webp" alt="Test call: a chat with a scripted agent, the step it's on, and the details it collected">
  221. </picture>
  222. <p><b>🧪 Test calls.</b> Chat or talk through your microphone. The steps light up as the call moves, and every collected detail shows on the right.</p>
  223. </td>
  224. <td width="50%" valign="top">
  225. <picture>
  226. <source media="(prefers-color-scheme: dark)" srcset="docs/images/integrations-dark.webp">
  227. <img src="docs/images/integrations-light.webp" alt="Integrations: model providers as tiles, with the one in use highlighted">
  228. </picture>
  229. <p><b>🧠 20 model providers.</b> Pick one, paste a key, press Test connection. Speech runs in the cloud or free on your own computer.</p>
  230. </td>
  231. </tr>
  232. <tr>
  233. <td valign="top">
  234. <picture>
  235. <source media="(prefers-color-scheme: dark)" srcset="docs/images/connections-dark.webp">
  236. <img src="docs/images/connections-light.webp" alt="Phone numbers and SIP, and contact centers marked coming soon">
  237. </picture>
  238. <p><b>📞 Phone calls and SIP.</b> Twilio, Telnyx, or your own SIP trunk, PBX or SBC through jambonz. Genesys, Avaya, Cisco and more are coming.</p>
  239. </td>
  240. <td valign="top">
  241. <picture>
  242. <source media="(prefers-color-scheme: dark)" srcset="docs/images/data-dark.webp">
  243. <img src="docs/images/data-light.webp" alt="Data: an appointments collection with sample records">
  244. </picture>
  245. <p><b>🗄️ Data of its own.</b> Collections agents look up and save during calls, and document search over your FAQs, all on your machine.</p>
  246. </td>
  247. </tr>
  248. </table>
  249. <sub>Screenshots use the bundled sample agents and made-up data.</sub>
  250. ## 🏠 Where it runs
  251. | | Status | |
  252. |---|---|---|
  253. | **Local and on-prem** | ✅ Available | One Docker container, or the engine natively, on any machine you control |
  254. | **Google Cloud** | 🔜 Coming soon | A one-command deployment to your own Google Cloud project |
  255. ## 🧠 Models
  256. Pick one on the Integrations page; each starts on a sensible default model you
  257. can change, and **Test connection** checks the key.
  258. | Provider | Key |
  259. |---|---|
  260. | Claude (Anthropic) | `ANTHROPIC_API_KEY` |
  261. | OpenAI, or any OpenAI-compatible endpoint | `OPENAI_API_KEY`, `VPE_LLM_BASE_URL` |
  262. | Gemini | `GEMINI_API_KEY` |
  263. | Gemini on Vertex AI | `GOOGLE_CLOUD_PROJECT` plus a service account key or this computer's Google credentials |
  264. | Azure OpenAI | `AZURE_OPENAI_API_KEY`, `AZURE_OPENAI_ENDPOINT` |
  265. | Amazon Bedrock | AWS keys, or this computer's AWS credentials |
  266. | Groq, Mistral, DeepSeek, Grok (xAI), Together AI, Fireworks AI, OpenRouter, Cerebras, NVIDIA NIM, Qwen, SambaNova, Nebius, Novita AI | each provider's API key |
  267. | Ollama | none: open models on this computer or your own server |
  268. ## 📞 Phone calls and SIP
  269. | | |
  270. |---|---|
  271. | Twilio | a Twilio number rings your agent; outbound calls from triggers |
  272. | Telnyx | the same, through a TeXML application |
  273. | SIP through [jambonz](https://jambonz.org) | any SIP trunk, PBX or session border controller: point a jambonz application at the engine |
  274. The Phone numbers page has the webhook to copy and the steps for each.
  275. Details: [engine/README.md](engine/README.md#phone-calls-twilio-telnyx-or-sip).
  276. ### 🏢 Contact centers (coming soon)
  277. Genesys Cloud, Avaya, Cisco Webex Contact Center, Cisco UCCE and PCCE, Amazon
  278. Connect, NICE CXone, Five9 and Twilio Flex show on the Integrations page as
  279. coming soon. Until each lands, most of them can already send calls to your
  280. agents over a SIP trunk through jambonz. The plan for each:
  281. [docs/ROADMAP.md](docs/ROADMAP.md#contact-centers-coming-soon).
  282. ## ⚙️ How it works
  283. ```mermaid
  284. flowchart LR
  285. you((You)) --> app[Lampline app<br/>builder, tests, integrations :7860]
  286. app --> engine[Engine<br/>Pipecat pipelines]
  287. phone((Callers)) --> carriers[Twilio · Telnyx ·<br/>SIP via jambonz]
  288. carriers --> engine
  289. engine --> agents[(agents/<br/>YAML + Markdown)]
  290. engine --> data[(Records and<br/>knowledge, SQLite)]
  291. engine --> models{{Model and speech<br/>cloud APIs or local<br/>Whisper · Moonshine · Parakeet · Kokoro · Qwen3-TTS · Ollama}}
  292. engine -- tools --> hooks[Your APIs and<br/>MCP servers]
  293. ```
  294. The engine reads the agent folders on every call, so a change is live on the
  295. next call. Settings and keys live in a `chmod 600` file outside the repo.
  296. ## 💰 What a voice-agent call actually costs
  297. Monthly totals for 4-minute calls at list prices (October 2026). Staff time is
  298. not included. Reproduce or change any assumption with `python3 costs/cost_model.py`.
  299. | Deployment | 10k calls/mo | 50k calls/mo | 200k calls/mo |
  300. |---|---|---|---|
  301. | CX Agent Studio (list price, baseline) | $5,000 | $25,000 | $100,000 |
  302. | On-prem server (no GPU) + Google model APIs | $1,384 | $6,235 | $24,426 |
  303. | On-prem GPU servers + self-hosted models | $681 | $1,363 | $3,748 |
  304. | Cloud Run + Google model APIs (coming soon) | $1,423 | $6,356 | $24,908 |
  305. | GKE production + mixed vendors: Deepgram, Claude Haiku, Fish Audio (coming soon) | $1,105 | $3,612 | $13,037 |
  306. | GKE production + self-hosted models on L4 GPUs, 3-yr CUD (coming soon) | $1,054 | $1,675 | $3,888 |
  307. What the numbers say:
  308. - 📉 **Hosting is the small part.** Model APIs are almost the whole bill, so the model choice matters far more than local vs. cloud.
  309. - 🖥️ **Self-hosted models are where the big savings are,** but they rest on two assumptions nobody has published good data for: concurrent calls per GPU, and whether an 8B open model is good enough for your flows. [Help measure them](CONTRIBUTING.md).
  310. Every row, per-call figures and assumptions: [costs/COSTS.md](costs/COSTS.md).
  311. ## 🧩 Agents as code (`vk`)
  312. Build and change agents from a terminal or an AI coding agent as well as the
  313. builder. Each agent is a folder: `agent.yaml` for the graph and one Markdown
  314. file per prompt, so changes are reviewable in git.
  315. ```bash
  316. pipx install lampline # installs the `vk` command (from a clone: pip install -e cli)
  317. vk templates install front-desk # start from a ready-made agent
  318. vk lint agents/front-desk # graph checks + a scan for real-looking patient data
  319. vk test agents/front-desk # runs tests/*.yaml conversations on the engine
  320. vk import dfcx export.zip # bring an agent over from Dialogflow CX or CX Agent Studio
  321. vk skills install --target claude # the four coding-agent skills (or install the Claude Code plugin, above)
  322. ```
  323. Details in [cli/README.md](cli/README.md). The engine itself, its builder,
  324. tools, data and knowledge: [engine/README.md](engine/README.md).
  325. ## 🗺️ Roadmap
  326. - [x] The engine on Pipecat: builder, text chat, browser calls, phone calls ([engine/README.md](engine/README.md))
  327. - [x] Agents as code: `vk` lint, conversation tests, templates, coding-agent skills ([cli/README.md](cli/README.md))
  328. - [x] Scripted pages and flows mixed with LLM steps ([engine/README.md](engine/README.md#scripted-pages-and-flows))
  329. - [x] Dialogflow CX and CX Agent Studio import and export ([cli/README.md](cli/README.md#import))
  330. - [x] Records and local RAG of its own ([engine/README.md](engine/README.md#data-and-knowledge))
  331. - [x] One Docker container for local and on-prem
  332. - [x] 20 model providers; SIP through jambonz
  333. - [x] Transfer to a person with the call's details as SIP headers ([engine/README.md](engine/README.md#transfer-to-a-person))
  334. - [ ] **Google Cloud deployment** (coming soon)
  335. - [ ] **Contact centers:** Genesys Cloud, Avaya, Cisco Webex Contact Center, Cisco UCCE, Amazon Connect, NICE CXone, Five9, Twilio Flex (coming soon)
  336. - [x] Users and roles, and an HTTPS guide ([docs/HTTPS.md](docs/HTTPS.md))
  337. - [x] Test a phone agent for free from your laptop ([docs/PHONE-TESTING.md](docs/PHONE-TESTING.md))
  338. - [x] Evals: real phone calls through Google Voice, played by a Test Caller and graded ([docs/EVALS.md](docs/EVALS.md))
  339. - [x] Import Hammer, Cyara or spreadsheet test scripts as eval test cases ([docs/EVALS.md](docs/EVALS.md#import-test-scripts))
  340. - [x] Improve an agent from its failing test cases ([docs/EVALS.md](docs/EVALS.md#improve-an-agent))
  341. - [ ] Evals placed by Lampline through Twilio, for unattended runs
  342. - [x] Call history and optional recordings, on this computer ([engine/README.md](engine/README.md#call-history))
  343. - [ ] Published benchmarks
  344. The full list: [docs/ROADMAP.md](docs/ROADMAP.md).
  345. ## ✅ Tested
  346. CI runs on every push and pull request to `dev`, `qa` and `prod`:
  347. - the `vk` CLI test suite
  348. - the engine test suite: flows, extraction, text chat, evals, every model provider's setup, Twilio, Telnyx and jambonz calls, the voice pipeline and a real WebRTC handshake, on fake models
  349. - the web app builds, and the committed build matches its source
  350. - the Docker image builds, starts as a non-root user, answers health and refuses requests without its key
  351. - shellcheck, and a check that `COSTS.md` matches the cost model
  352. - a scan of every commit for keys and tokens (gitleaks)
  353. Issues and PRs welcome (against `dev`; [how dev, qa and prod work](docs/BRANCHES.md)). See [CONTRIBUTING.md](CONTRIBUTING.md), [SECURITY.md](SECURITY.md), the [code of conduct](CODE_OF_CONDUCT.md) and the [changelog](CHANGELOG.md).
  354. ## 🙌 Credit and citation
  355. Lampline is created and maintained by **Rohan Salvi**. If you use it in a
  356. project, product, paper or talk, please credit "Lampline by Rohan Salvi" and
  357. link this repository. GitHub's **Cite this repository** button (from
  358. [CITATION.cff](CITATION.cff)) gives APA and BibTeX:
  359. ```bibtex
  360. @software{salvi_lampline_2026,
  361. author = {Salvi, Rohan},
  362. title = {Lampline: open-source, self-hosted voice agents},
  363. year = {2026},
  364. version = {0.3.0},
  365. url = {https://github.com/C4rohan/lampline}
  366. }
  367. ```
  368. If you redistribute Lampline or build on it, the Apache-2.0 licence requires
  369. you to keep the [NOTICE](NOTICE) file, which carries this attribution.
  370. ## 📄 License
  371. Apache-2.0. The engine uses Pipecat (BSD-2-Clause), installed from PyPI;
  372. other credits are in [NOTICE](NOTICE). Not affiliated with Google or any
  373. provider named here. Prices are list-price estimates that change often; check
  374. them before budgeting.

Ratings and comments

Sign in to rate this skill.

Sign in

Free to use. If it’s helped, pass it on.