docs/ai.md: where your models answer (Chat and Vibe, Live's AI viewers, any of your computers, the SDK) (#77)

ddiff --git a/docs/ai.md b/docs/ai.md
iindex 1cd0d19..e41ed42 100644
---- a/docs/ai.md
++++ b/docs/ai.md
@@@ -53,6 +53,22 @@ computer's models** for the machine on openvibe.bot/machines.
  | `max_tokens` | `2048` | the most one reply may be (at most 32768) |
  | `timeout_s` | `180` | how long one reply may take (at most 900) |
  
++## Where your models answer
++
++Once "Let my AI use this computer's models" is on for the machine, on openvibe.bot/machines:
++
++- **OpenVibe.Chat.** Under "Your AI key" in your settings, choose "A model on my own computer". The machine card's
++  "Use these models in OpenVibe.Chat" fills it in. Your Vibe (tick "Vibe thinks with it too") and the AI members you
++  set up in rooms then think with it.
++- **OpenVibe.Live.** A streamer's AI chat viewers: Dashboard, Chat AI, AI Chat Viewers, Key & Budget, AI source "A
++  model on my own computer (free)".
++- **Any of your computers.** Choose "any" instead of one machine. OpenVibe asks the fastest of yours that is on and
++  has the model (by `ai_tokens_per_s`), then the next when one is busy. It is never anyone else's machine.
++- **Your own code.** `openvibe-sdk/bot`: `bot.machines.chat(id, { model, messages })` asks one of the machine's
++  models as you (SDK 0.67.0).
++
++When the machine is off or its model server is down, the request fails and says so. Nothing else answers in its place.
++
  ## The command
  
  `ai` `{"model", "messages": [{"role": "system"|"user"|"assistant", "content"}], "max_tokens"?, "temperature"?}` →