Vibium v26.11 ships: real-device voice, Run/Check autonomous browser tasks
hugs · x · 2026-11-18
hugs releases Vibium v26.11, the project's biggest release yet — 245 merged PRs and 430 commits since 26.8, double its size.
Key capabilities:
- Run: vibium run "goal" executes plain-English browser tasks in your existing session, returning COMPLETED/NOTCOMPLETED with evidence; available in JS/Python/Java clients and as an MCP tool.
- Check: vibium check "claim" verifies whether something is true (PASS/FAIL).
- Find: locate elements from a description or picture.
- Say/Listen: built-in AI TTS and STT so an agent can talk to a page and hear it talk back.
- Real mobile devices: first-class support for phones/tablets via Roadie and Vibium Station, aimed at builders and testers of voice-powered mobile apps.
The author had Claude produce a highlight video, which used "vibium say" for the voiceover and "vibium listen" to verify it.
More from coding & agent
- Astra 6.2 solves in 2 hours a GPU bug that Fable 5.2 and Astra 6.1 both failed — i_dg23 · 2026-11-18
- GPT-6.2 Sol builds fight game in 12 minutes, wins Wikipedia computer-use test at 39s — iamfakhrealam · 2026-11-18
- Fire sim head-to-head: Sol done in 12 min for $0.73, Opus 5.5 took 85 min and $18.40 — iamfakhrealam · 2026-11-18
- Peter Yang finds Gemini 6.2 lazy on Medium effort, mulls switching to Hard — petergyang · 2026-11-18
- Reddit: how to pick an enterprise AI agent platform for a read-only procurement assistant — CreditGreen992 · 2026-11-18
- 10 open-source GitHub repos for running a whole team of AI agents — victor_explore · 2026-11-18