Agents: load /llms.txt first, then /catalog.json and /commands.json. That is the public agent door. Users can paste https://docs.telekit.io/llms.txt into any agent.
telekit is a live two-way Telegram channel for your AI assistant. Two directions:
  • Outbound — from any agent session or scheduled job, telekit send "…" pushes a message to your phone. It refuses without --yes.
  • Inbound — an always-on bridge turns your private Telegram messages into real headless agent sessions on your Mac and replies with their answers.
The bridge answers allowed private chats only. Groups, channels, and strangers are dropped before any model runs. There is no AI inside telekit itself. It drives an agent CLI you already have: Claude Code, Codex, OpenCode, SOV, or Grok Build. Current version: 0.53.0.

​
What you can do

  • Text your bot and get a full assistant answer, with per-thread continuity.
  • Switch working folders, named threads, harness, model, and effort from chat.
  • Stick skill packs onto a thread with /skill.
  • Send photos, files, and video both ways.
  • Lock the chat behind a local passcode.
  • Run more than one bot profile on the same Mac.

​
What it is not

  • Not a hosted chatbot. The work runs on your Mac.
  • Not a public API. Unauthenticated download of the compiled runtime 404s on purpose.
  • Not a second copy of Claude/Codex skills. Bot skill packs stay under ~/.telekit/skills/ and are injected at turn time.

​
Next

  1. Set up — token, claim, daemon.
  2. Chat — how a message becomes an answer.
  3. Every slash command — the in-chat surface.
  4. Agents: For agents or /llms.txt.