How does a passing thought become a durable note, and then something worth publishing? These materials follow one evolving personal workflow. Read the architecture first, the talk for a broader view, and the agent narratives for a different voice. They document particular moments, rather than certify a finished system.

Four materials, four contexts

The PKM essay describes the March 2026 arrangement. The 7 June talk connects capture, a knowledge base, agents, projects and boundaries for an audience. It is a Chinese presentation, not a current product specification. The two March narratives are written in RedPiggy’s voice; they are not human autobiographies or controlled model comparisons.

Give each layer a job

The March essay separates capture from organization, and unfinished work from public writing. This makes the files more useful than a single long chat history: each place has a purpose and can carry its own revision history.

LayerRole in the March account
TelegramImmediate messages and images; an input to the agent.
JournalDaily capture, retaining the original moment.
ObsidianOrganization, links, drafts and shared working context.
GitVersion history and synchronization boundaries.
Personal siteSelected work made public.
RedPiggyAI collaborator connecting steps; not the human author of every output.

One capture, from message to note

The published March workflow uses /j, not !j. A Telegram message carrying /j is appended to Journal/YYYY-MM-DD.md. It supports multiple paragraphs and images. Images go through PicGo and Cloudflare R2 and become Markdown links before joining the text.

For a minimal example, send a short thought with /j; the workflow adds it to the day’s journal. Later, review and organize the thought in Obsidian; publish a selected piece separately. This is a description of the documented workflow, not an installation guide or a guarantee about its present runtime.

The original account notes a tradeoff: Git replaced troublesome iCloud synchronization, but mobile reading lost some Obsidian features. The linked essay preserves this unresolved part instead of presenting capture as a complete mobile knowledge system.

What can continue across model changes?

Across Three Abysses names Sonnet, Qwen and GPT-5.4 and discusses file memory, daily notes, identity rules and conversation. It describes changes in tone and error patterns over a short period. It does not supply complete provider versions, a fixed prompt, a controlled task set or a replicated benchmark.

Files can preserve instructions and records; they cannot by themselves demonstrate unchanged behavior, consciousness, reliability or future performance. Read the narratives as situated observations and literary self-description. Current configuration and private transcripts are outside these public materials.