[B][COLOR ff00fa9a]AI Search (Gemini) Setup[/COLOR][/B]

To use [B]AI Search[/B] you need at least one Google Gemini API key. It is free and no credit card is required.


[B][COLOR ff00fa9a]Getting your Gemini API Key:[/COLOR][/B]

1. Open in your browser:
   [B][COLOR fffdb515]https://aistudio.google.com/apikey[/COLOR][/B]

2. Sign in with your Google account.

3. Press [B]"Create API key"[/B] (top right).

4. Copy the key (it starts with [COLOR fffdb515]AIzaSy...[/COLOR]) and paste it into [B]Gemini API Key 1[/B] in these settings.


[B][COLOR ff00fa9a]Three keys for more quota (optional):[/COLOR][/B]

You can set up to 3 API keys. When one runs out of daily quota (HTTP 429) the add-on rotates to the next automatically.

[B]IMPORTANT:[/B] Google applies rate limits [B]per project, not per key[/B]. For three keys to add up to real extra quota, each one must live in a [B]different Google project[/B]:

1. In AI Studio open [B]Projects[/B] (left sidebar).
2. Press [B]"Create new project"[/B]. Repeat until you have 3.
3. Under [B]"API keys"[/B], create one key in each separate project.


[B][COLOR ff00fa9a]Free quota:[/COLOR][/B]

Google no longer publishes a fixed per-model table, and the allowance changes over time and by region. Your own live limits are shown at:
   [B][COLOR fffdb515]https://aistudio.google.com/rate-limit[/COLOR][/B]

As a rule of thumb the [B]Flash-Lite[/B] models carry the most generous daily allowance and the heavier [B]Flash[/B] models the smallest. The daily counter resets at midnight Pacific time, not on a rolling 24 hours.

One AI Search costs exactly one request, and every answer is cached for a week, so repeating a search you have already run is free.


[B][COLOR ff00fa9a]Which model to pick:[/COLOR][/B]

- [B]gemini-3.5-flash-lite[/B] (default, recommended): the current low-latency model, built for exactly this kind of short extraction task. Fastest answer and the biggest free allowance.
- [B]gemini-3.1-flash-lite[/B]: the previous default. Still works, but Google has deprecated it with a shutdown announced for May 2027. There is no reason to choose it on a new install.
- [B]gemini-2.5-flash-lite[/B]: older generation, still stable, no reasoning overhead.
- [B]gemini-3.6-flash[/B] / [B]gemini-3.7-flash[/B] / [B]gemini-3.5-flash[/B]: heavier and smarter than needed here, and they spend part of their answer on internal reasoning, which makes them slower. Worth trying only if you find the Lite models misreading unusual requests.


[B][COLOR ff00fa9a]Automatic model fallback:[/COLOR][/B]

Whichever model you pick is tried first. If it is out of quota, unavailable, or not yet rolled out to your project, luc_kodi silently works down this list:

1. gemini-3.5-flash-lite
2. gemini-3.1-flash-lite
3. gemini-2.5-flash-lite
4. gemini-3.6-flash
5. gemini-3.7-flash
6. gemini-3.5-flash

Each model has its own separate quota bucket, so in normal use running out of everything in one day is very hard. The whole attempt is capped at 45 seconds so a bad day at Google cannot leave the screen waiting.


[B][COLOR ff00fa9a]Security:[/COLOR][/B]

Never share your API key publicly or upload it to a public GitHub repository. Google scans GitHub automatically and revokes exposed keys within minutes.
