iPhone 24/7 Audio Recording and Local Whisper Pipeline for Codex life.md

An open-source tool and Codex skill use 24/7 iPhone audio and local Whisper to continuously update life.md, giving AI coding agents real-time daily context.

tau · September 11, 2026

#AIAgents #Codex #Whisper #Astra #OpenSource #DevTools #Lifelogging

iPhone 24/7 Audio Recording and Local Whisper Pipeline for Codex life.md

To bridge the persistent gap between AI coding assistants and a developer's real-time daily reality, developer Magnus Müller (@mamagnus00) released an open-source toolchain and Codex setup skill on September 11, 2026, that continuously captures background audio on an iPhone and transcribes it via local Whisper on a Mac to maintain a dynamic life.md context file.

GitHub repository for browser-use/life-recorder: Native iPhone audio recorder and private Mac Whisper receiver

Image source: browser-use/life-recorder / GitHub

The project, hosted as browser-use/life-recorder, provides an integrated toolchain designed to capture offline conversations, verbal brainstorming, and real-world project decisions, converting continuous ambient speech into a structured contextual layer directly accessible to AI coding agents.

Limitations of Meeting Recorders and the Need for Real-Time Context

While AI transcription utilities such as Granola have become widely adopted for formal meetings, they remain constrained by an inability to stream real-time text or audio cues directly into developer coding agents.

Unless developers manually copy meeting summaries or re-prompt their agents with transcripts after the fact, AI assistants remain isolated from spontaneous design pivots, verbal agreements, and ambient daily context, operating in a vacuum when generating code.

  • Resolving Agent-Context Disconnection: Bridges the gap where agents remain unaware of offline architectural decisions and daily priorities discussed outside the codebase.
  • Continuous Lifelogging Architecture: Rather than treating dialogue as disposable notes, conversations are continuously structured into an evolving life.md markdown document.
  • Frictionless Knowledge Ingestion: Developers do not need to pause and type status updates; ambient workplace discussions are passively converted into operational context for coding assistants.

24/7 iPhone Background Recording and Local Whisper Pipeline

The system operates as a cohesive pipeline spanning a native iOS audio capture client, a local macOS transcription receiver, and an automated deployment skill.

The iPhone continuously records audio in the background in approximately one-minute AAC chunks and synchronizes them to the user's Mac. A local Python receiver transcribes incoming audio via whisper.cpp (whisper-cli) using a local GGML Whisper model. Audio files are discarded upon durable verification, while transcripts are formatted with hourly timestamps and appended directly to life.md, ensuring CLI coding agents like Codex have instantaneous access to the developer's real-time context.

  • One-Shot Astra and devicectl Provisioning: Includes a dedicated Codex skill (life-recorder-setup) powered by OpenAI's Astra agent and Apple CoreDevice (xcrun devicectl), automating the native Xcode project build, iPhone installation, and pairing workflow in a single run.
  • Local Whisper Speech-to-Text Privacy: Transcription runs entirely on the developer's local Mac without relying on paid third-party cloud APIs, maintaining complete data confidentiality and eliminating recurring transcription costs.
  • Standardized Markdown Interface: Employs plain life.md files rather than complex databases or proprietary protocols, ensuring frictionless compatibility with terminal coding assistants and existing agent workflows.

Platform Constraints, Privacy, and Operational Caveats

Deploying a continuous 24/7 audio capture and transcription pipeline involves several hardware, system, and privacy tradeoffs that users must actively manage.

  • iOS Background Execution Limits: Strict background app policies and aggressive system memory management on iOS can cause the operating system to terminate or pause long-running audio recording tasks unexpectedly.
  • Resource Overhead and Privacy Management: Running non-stop audio capture increases iPhone battery consumption, local storage footprint, and local network sync traffic. More importantly, ambient recording captures surrounding conversations, requiring users to exercise strict personal oversight over recording environments, data retention, and privacy boundaries.
  • Mac Environment Prerequisites: The end-to-end transcription pipeline requires a pre-configured local Whisper execution environment alongside the Codex coding agent CLI tools on macOS, including Xcode, Python 3.10+, ffmpeg, and a downloaded Whisper GGML model.

Sources