One tool, nine doors — trip2g

Your vault becomes a website as it is: wikilinks turn into links and backlinks, and a note keeps growing after it is published. You write in Obsidian, sync, and the page is live on your own domain. There is no export step, so the site is the working vault, rough parts included.

A post is an ordinary note: add two properties and it goes out to your Telegram channel on schedule. trip2g converts the formatting, keeps the queue and carries later edits into the post that is already live. The channel and the site grow from the same notes.

Like Boosty or Patreon, except the reader pays for a living knowledge base, not a feed of posts. Part of the vault is open to everyone, the rest to subscribers who pay through Patreon, Boosty, a paid Telegram group or crypto. A subscriber's AI agent connects to the same base over MCP in subscription-only mode: instead of transcribing hundreds of hours of an expert's videos, you plug their always-current base into your agent.

Claude Code, Codex or any MCP agent starts every session blank. Point it at trip2g and what it learns stays as markdown notes: the next session finds them by search and reads only the section it needs. The memory is files you can read, diff and fix by hand, the same on every machine and shared with a team. memcli boots a local server in one command.

Your notes and docs answer agents over MCP: search returns the matching section with its place in the outline, and the agent reads just that part. Instructions written as an ordinary note reach the agent when it connects, and named entry points give one base several roles. Expose the whole base or one subgraph; the base's own access rules apply.

Everyone keeps their own base and writes only there. To reach a neighbour's knowledge you don't pile everything into one shared wiki or wire your agent to ten bases: your own hub routes the query. A neighbour's base is one note in your hub with its address; federated_search sends the question there, or to every base at once, and each answer keeps the name of the base it came from. A data mesh for knowledge, and a circle of friends is the same, only smaller.

A bot that hands a person the note they asked about instead of a whole "about me" page. What works today: a Canvas file drives a navigation bot, and the reader moves through your notes by buttons right in the chat. The bot that asks first and learns from the questions it could not answer is in development.

A recorder such as Krisp captures the call, and an LLM pipeline turns it into linked notes: the verbatim transcript, a call note whose topics quote it, concept notes for people and terms, a daily note with action checkboxes. Not a summary plus a chat bot, but a graph that grows with every call. Next to it, agents keep the team's journal and report to the team on their own.

Content lives in markdown files in your own folder, not in the CMS's database. The server renders them at runtime: one Go binary on one SQLite file, no separate database server, no build step. Edit a file and the page changes; leave and you take every file with you. A whole base can even ship as one container image.