The Walking Dev
My first open-source project: an autonomous agent that produces a personal podcast every night, listened to in the morning during a 10,000-step walk. News, project progress, emails and calendar, synthesized into voice by AI. Designed clone-and-run from day one.
- Year
- 2026
- Status
- Open source
A personal product turned open source: get a 10-to-20-minute audio brief every morning, blunt and to the point, covering curated news on chosen topics, project and goal progress, a digest of useful emails and the day's agenda. The goal: cut the friction of starting the day, and ship a product anyone can clone and run for themselves.
Full design and development, solo, through to the open-source release: a six-adapter interchangeable architecture, an automated nightly pipeline, script writing by headless Claude Code, voice synthesis, feed hosting and a conversational bot. Complete code sanitization, a test suite, CI and English docs for clone-and-run.
- Daily audio brief generated overnight (news, projects and goals, daily challenge, emails, agenda) and published to a private podcast feed
- Script written by headless Claude Code, fed by curated RSS news, the notes vault, emails and calendar
- AI voice synthesis (OmniVoice) with a cloned voice, around 40× real time on GPU
- Conversational Telegram bot: onboarding, adaptive evening questions and clickable daily tasks
- Fully local, privacy-first mode: local LLM writer (Ollama, LM Studio…), CLI channel, toggleable news, zero outbound requests
- Six interchangeable adapters: messaging, knowledge provider, state store, script writer, audio hosting and TTS engine
- First published open-source project (MIT license, v0.1.0 release)
- Runs on its own: scheduled tasks (nightly generation and persistent bot)
- Clone-and-run architecture: six adapters and a fully local privacy mode
- Personal data and psychology moved out of the source code, offline test suite and green CI