NSNeeraj Sharmainaxionry.hashnode.dev·1d ago · 3 min readOur voice agent costs 2.5¢ a minute. Here's the whole bill.We run an AI interviewer that talks to candidates over phone calls. It started on a managed voice platform at roughly 10¢ a minute. The same interviews now run on a custom LiveKit pipeline at about 2.00
NSNeeraj Sharmainaxionry.hashnode.dev·1d ago · 3 min readShowing live voice agent transcripts in a Flutter app with LiveKitIf you build a voice agent with LiveKit Agents and a Flutter frontend, at some point you want the conversation on screen: what the user said and what the agent is saying, updating live. The docs cover00
YLYaoshen Luoinluoyaoshen.hashnode.dev·4d ago · 7 min readBenchmarking Real Work - Case 01: How I Built a Voice Agent Benchmark from Real Customer FailuresCustomer complaints are not the problem definition; they are the signal. This post captures a real-world case of agent benchmarking: how I built v1 of our benchmark from real customer failure logs—wi11L
MUMohammed Usmaniinmohammed-usmani.hashnode.dev·4d ago · 11 min readFrom 2.44 s to 1.81 s: working through my voice agent's latency list, and the fast model that went silentTL;DR Part 1 ended with a list of things to try. I tried all of them on the same phone agent, now running a recruiter's pre-screening call. At the agent, the median reply went from 2.44 s to 1.81 s 00
ETEcosmob Technologyinecosmobvoipexpert.hashnode.dev·5d ago · 16 min readBuilding Voice AI for Real Phone Calls: 5 Global Platforms and the Telecom Architecture Behind ThemIf you've ever built a voice AI demo in a browser, you probably know the experience: The response feels instant. The voice sounds natural. Interruptions work reasonably well. Then you connect the same00
LLokutorinlokutor.hashnode.dev·6d ago · 4 min readVoice agents on CPU vs the GPU incumbents: latency, cost, and a deliberate comparisonTwo numbers frame this post. On Lokutor, a production voice agent starts speaking in 139 ms at the median (TTS streaming time to first audio, one CPU thread), and a full minute of a running agent cost00
LLokutorinlokutor.hashnode.dev·6d ago · 4 min readMigrate from ElevenLabs to a CPU voice API in about 20 linesMoving text-to-speech from ElevenLabs to Lokutor is a small diff: same install, same one call, same play-through-speakers result. What changes is what sits behind the call: a few-tens-of-millions-para00
LLokutorinlokutor.hashnode.dev·6d ago · 3 min readElevenLabs alternative for real-time voice agents: CPU-native, 2 cents a minuteElevenLabs is the default answer for synthetic voices, and for good reason: a large voice library, strong voice cloning, dubbing, and a broad creative suite. If you are producing narration, dubbing vi00
LLokutorinlokutor.hashnode.dev·6d ago · 5 min readVoice agent pricing at real volumes: the monthly bill from 500 to 100,000 minutesPer-minute list prices are the easiest numbers in voice AI to compare and the least useful. Nobody pays list price: you pay a monthly plan plus whatever your minutes spill over it, and the plan decide00
MUMohammed Usmaniinmohammed-usmani.hashnode.dev·Sep 26 · 11 min readLiveKit turn detection on real phone calls: why a 2-second average hid a 3.6-second stallTL;DR I built a voice agent you can call from a real phone (SIP through LiveKit Cloud) and measured every turn instead of trusting averages. On a test call the average reply took about 2 seconds, bu00