JKinco Listen (Open Edition) is a local-first meeting-minutes workbench that carries a recording through transcription, meeting-type detection, structured minutes and DOCX or PDF export without a cloud call or an API key. Speech recognition runs on FunASR's paraformer-zh with voice-activity detection, punctuation restoration and a domain-lexicon correction pass; the text stages — scene detection, minutes generation, quality review and cross-meeting Q&A — run against a model the operator hosts, by default qwen2.5:7b-instruct through Ollama, or any OpenAI-compatible endpoint. Meetings, transcripts, minutes and human edits stay in a local SQLite file. Scene detection is deliberately two-tier: a rule-based evidence gate runs first and has the final say, so the model cannot classify a meeting as a formal construction site review on generic words like project, schedule or quality unless the accountable parties and a production-control agenda actually appear in the transcript. Five meeting types are covered — construction review, general minutes, personal assistant, interview record and customer visit — plus custom template uploads, exported with the original corporate layout preserved. The stack is FastAPI and React with TypeScript on Python 3.12, and docker compose up brings the web UI, backend, Ollama and the model pull up in one command; the project reports 961 automated tests and a CI scan that fails the build if cloud-model references appear in the repository. The accuracy figure it publishes is narrow and honestly framed: across ten noisy Chinese meeting clips the local ASR alone recognised 31 of 36 domain terms and 36 of 36 with correction enabled, on synthetic audio the README itself calls cleaner than a real meeting room and therefore an upper bound. Other limits are worth knowing before adopting it: the ASR model and lexicon are tuned for Chinese, live captioning is experimental and off by default behind an environment flag, speaker diarization and multi-language support sit unchecked on the roadmap, and .env.example ships a default admin password on a container that listens on 0.0.0.0.