Autofill AI Ninja
Configuration
Side panel layout, persisted settings, and supported LLM providers.
Side panel reference
| Section | Controls |
|---|---|
| Run | Custom request textarea, Fill this page, status line |
| Progress | Live log of the current/last run, Clear |
| Connection | Provider, API key (Test / Save / Remove), remember & encrypt options, model |
| Behavior | Fill mode, language, auto-next, step/round/settle limits, empty-only, sensitive skip. Changes save automatically (API keys still use Save key). |
UI strings ship in English, German, Japanese, Spanish, Portuguese, and Bengali (_locales/en, de, ja, es, pt, bn).
Settings
Settings live in chrome.storage.local under aff_settings and are
sanitized on every read.
| Setting | Default | Meaning |
|---|---|---|
provider / baseUrl / model |
OpenAI / https://api.openai.com/v1 / gpt-4o-mini |
Selected provider, endpoint, and model |
fillMode |
hybrid |
hybrid, ai_only, or heuristics_only |
fillLanguage / fillLocaleOverride |
auto / empty |
Follow the page, or force a BCP-47 locale |
customRequest |
empty | Free-text user rules |
maxRounds |
24 (range 1–60) |
Chunk attempts per step |
settleMs |
120 (0–3000) |
Extra delay after writes, on top of mutation-based settling |
autoNextEnabled / autoNextMaxSteps |
true / 15 (cap 50) |
Automatic advancing |
fillEmptyOnly |
true |
Skip fields that already hold a value |
excludeSensitiveFields |
false |
Skip password and payment (cc-*) fields |
rememberKeyAcrossRestarts |
true |
Persist keys, or drop them on browser start |
encryptKeys |
false |
Encrypt stored keys at rest behind a passphrase |
LLM providers
gemini-3.1-flash) — it is faster and makes fewer fill mistakes than older
Flash models. For Groq, use Llama 3.3 70B Versatile. Step-by-step:
Tutorial → Get a free API key.
| Provider | Kind | Default model |
|---|---|---|
| OpenAI | OpenAI-compatible | gpt-4o-mini |
| Anthropic | Anthropic messages | claude-3-5-haiku-latest |
| Google AI Studio | OpenAI-compatible | gemini-3.1-flash |
| xAI | OpenAI-compatible | grok-3-mini |
| Groq Cloud | OpenAI-compatible | llama-3.3-70b-versatile |
| OpenRouter | OpenAI-compatible | openrouter/auto |
| Cerebras | OpenAI-compatible | llama-3.3-70b |
Anthropic uses its own adapter (POST /v1/messages,
x-api-key, top-level system, block-array content). Only the
selected provider is contacted. Form contents are not sent to another vendor.
Within a provider, a failed request walks the configured model, then the default, then the fallback list, and can step down output-token budgets when a request is rejected as too large. Auth failures stop the walk immediately.
See also installation, security and storage, and the tutorial.