Short on purpose. Most of what used to be on this list got automated instead — see
"Already handled for you" at the bottom so you know it's covered, not forgotten.
Last updated v9.10.446.
---
cd C:\KAI .\Start-KAI.ps1
Then hard-refresh the dashboard (Ctrl+Shift+R). Nothing from tonight is live until you do.
Or, from anywhere — including your phone (v9.10.446): Admin -> Server Control ->
⚠ Restart Full Fleet. Same thing exactly: it stops the whole stack, waits 5 seconds, runs
Start-KAI.ps1, and streams the live log into the modal so you can watch each service come
back. You do not need a terminal for this.
> Do not use ⟲ Restart Dashboard Server for this — that one restarts the :3001
> dashboard ONLY and deliberately leaves the bots running, so it cannot load bot, voice or
> shared/ changes. The panel now tells you which one you need.
Radio → Your library → Add music. Pick files off your PC (MP3, M4A, OGG, WAV, FLAC, up to 30MB each).
They upload to your own server and play with your phone screen off, with lock-screen controls.
Only you can do this because they're your files. No key, no account, no third party.
---
I could not remove this one. GOOGLE_API_KEY is already configured, but Google's API
requires a *search engine ID* that only your Google account can mint — there is no
programmatic way for me to create it, and no keyless source whose terms permit this use.
Get it at programmablesearchengine.google.com (turn ON "Search the entire web"), then one line in tools/oracle-discord/.env:
KAI_SEARCH_CX=<the search engine id>
Free, 100/day, no card. Until then the browser shows a card explaining exactly this and offers a working "search in my own browser" button — it isn't a dead end.
Calls use STUN only. When both people are on strict mobile networks there's no direct path and the call now says so plainly instead of ringing forever.
I left this to you because it's a spend/networking trade, not a command.
---
.\Install-FaceRecognition.ps1
One command — installs the packages, self-tests, and sets the flags. Skip unless you want it.
---
Groq → Cerebras → Gemini → xAI → local Ollama (the DJ and limited-mode bots use Groq → Cerebras → Ollama; KAI never uses an LLM at all). Per-model cooldowns stop a 429'd model being retried while cold. Start-KAI.ps1 already auto-starts Ollama as the terminal fallback, so there is a working local backend even with every cloud provider exhausted. Topping up quota was never required..\Run-GuestSuites.ps1. They spawn their own throwaway server on a random port — they don't need the fleet up and don't touch live data.<audio> element with Media Session controls._probe3.mjs and the state/credits/__test_*.json files affect nothing at runtime. Delete them whenever, or don't.---
scripts/measure-edge-tts-latency.mjs does not exist anywhere in the repo — I checked before asking you to run it. Nothing to run; the item was based on a bad reference.