AI Provider Setup - undefined Docs

AI Setup

FiavaionDictate can automatically clean up your dictated text — fixing grammar, smoothing out run-on sentences, and catching words the speech engine got wrong. This is optional. Dictation works fine without it.

When AI correction is on, each time you pause (or use the correct that command), your text is sent to whichever AI provider you have configured, and the corrected version is shown alongside the original so you can accept or reject it.

Your Four Options at a Glance

	Ollama (Local)	Google Gemini	Anthropic Claude	OpenAI GPT-4
Privacy	100% on your machine	Sent to Google	Sent to Anthropic	Sent to OpenAI
Cost	Free	Free tier available	Paid	Paid
Setup Time	5–15 min	2 min	2 min	2 min
Quality	Good (model-dependent)	Excellent	Excellent	Excellent
Works offline	Yes	No	No	No
Best for	Privacy, no ongoing cost	Most people starting out	Long-form, nuanced writing	General use

Our recommendation for most people: Start with Ollama if you’re comfortable installing software, or Google Gemini if you want the easiest setup. Gemini has a free API key with no credit card required.

Option 1: Ollama (Local, Free, Private)

Ollama is a free, open-source tool that lets you run AI language models on your own computer. When you use Ollama with FiavaionDictate, nothing you dictate is ever sent anywhere — the AI correction happens entirely on your machine.

Install Ollama

Download from ollama.com/download and run the installer. Ollama works on Windows, Mac, and Linux.

After installation, Ollama runs as a background service and listens at http://localhost:11434. You don’t need to open Ollama manually — it starts with your machine.

Choose a Model

Model	Download Size	Speed	Best For
Gemma 3.4B	~3 GB	Fast	Quick corrections, most everyday tasks
Llama 3.2 3B	~2 GB	Fastest	Older or lower-spec machines
Mistral 7B	~5 GB	Medium	Complex writing, long documents

Not sure? Start with Gemma 3.4B. It’s fast, runs on most modern machines with 8 GB RAM, and handles dictation correction very well.

Pull the Model

Open a terminal (Command Prompt on Windows, Terminal on Mac/Linux) and run:

ollama pull gemma3:4b

For Llama 3.2 instead:

ollama pull llama3.2:3b

For Mistral:

ollama pull mistral:7b

The first pull downloads the model — allow 5–15 minutes depending on your connection. After that, the model is cached locally and starts in under a second.

Verify FiavaionDictate Detects It

Once Ollama is running and you have at least one model pulled, open FiavaionDictate and look at the status bar at the bottom of the screen. The LOCAL badge should turn green with a checkmark.

Click the LOCAL toggle in the footer to activate Ollama as your AI provider. A dropdown will show your installed models — select the one you want to use.

Adding More Models Later

You can pull additional models at any time:

ollama pull <model-name>

FiavaionDictate detects new models automatically — no restart needed. The model dropdown refreshes when you open the AI settings.

To see all models you have installed:

ollama list

Option 2: Google Gemini (Free API Key)

Gemini is a great choice if you don’t want to install Ollama or if you’re on a machine without much RAM. Google provides a genuinely free API key — no credit card required, no subscription.

Why Gemini First

The free tier is legitimately useful, not a bait-and-switch. Flash gives you 250 requests per day and Flash-Lite is unlimited. For typical dictation use (correcting a few documents per day), you will not hit the limits.

Get Your Free API Key

Go to aistudio.google.com
Sign in with any Google account
Click Get API Key in the left-hand sidebar
Click Create API Key
Copy the key (it starts with AIza...)

That’s it. You’re never automatically charged. Google Cloud billing is a completely separate system that you’d have to explicitly set up — the AI Studio key doesn’t connect to billing by default.

Available Models and Limits

Model	Requests/Day	Requests/Min	Best For
Gemini 2.5 Flash	250	10	Everyday corrections, good balance
Gemini 2.5 Flash-Lite	Unlimited	30	High-volume use, fast responses
Gemini 2.5 Pro	100	5	Complex or lengthy writing tasks

Add the Key to FiavaionDictate

Click the CLOUD toggle in the footer
Select Google from the provider dropdown
Paste your key into the API key field
Click Test to confirm it works
Choose your preferred model from the model dropdown

The key is immediately encrypted and stored in your browser. You won’t need to enter it again.

Option 3: Anthropic Claude (Paid)

Claude excels at nuanced writing tasks — understanding context, matching tone, and making intelligent corrections rather than just fixing surface-level grammar. It’s a good choice for long-form content, academic writing, or anywhere where the quality of corrections really matters.

Get an API Key

Go to console.anthropic.com
Create an account (requires email verification)
Add credit to your account under Billing
Go to API Keys in the dashboard
Click Create Key, give it a name, and copy it

Available Models

Model	Speed	Cost	Best For
Claude Haiku 3.5	Fastest	Cheapest	Quick corrections, high volume
Claude Sonnet 4.5	Medium	Moderate	Balanced quality and cost
Claude Opus 4	Slowest	Most expensive	Long, complex documents

For dictation correction, Haiku is usually the right choice — it’s fast, inexpensive, and more than capable for grammar and clarity tasks. Reserve Sonnet or Opus for when you’re using AI commands like improve that or expand that on important documents.

Add the Key to FiavaionDictate

Click the CLOUD toggle in the footer
Select Anthropic from the provider dropdown
Paste your key and click Test
Select your preferred model

Option 4: OpenAI (Paid)

OpenAI’s models are capable, widely documented, and integrate well. GPT-4o Mini is a strong choice for cost-effective corrections.

Get an API Key

Go to platform.openai.com
Create an account and add billing
Go to API Keys in the left sidebar
Click Create new secret key, copy it

Available Models

Model	Speed	Cost	Best For
GPT-4o Mini	Fast	Low	Everyday corrections
GPT-4o	Medium	Moderate	Better reasoning, complex edits
GPT-4 Turbo	Slower	Higher	Long context, complex documents

Add the Key to FiavaionDictate

Click the CLOUD toggle in the footer
Select OpenAI from the provider dropdown
Paste your key and click Test
Select your preferred model

Switching Between Providers

You can have multiple providers configured and switch between them at any time — your keys are remembered per-provider.

Footer toggle: Click LOCAL or CLOUD to switch between Ollama and cloud AI. The badge shows which is active.

Provider dropdown: When CLOUD is active, the dropdown lets you switch between Google, Anthropic, and OpenAI. Switching is instant — the next correction request uses the newly selected provider.

Model dropdown: Each provider has its own model selection, remembered separately.

Security: How Your API Keys Are Stored

This is the question everyone should ask before pasting an API key into any app.

The short version

Your keys are encrypted in your browser and never leave your machine in any readable form. Fiavaion has no servers that receive your keys.

The technical details

When you paste an API key and click Save, FiavaionDictate:

Generates a random 16-byte salt using crypto.getRandomValues()
Derives a 256-bit encryption key from your browser fingerprint using PBKDF2 (SHA-256, 100,000 iterations)
Encrypts your API key using AES-GCM with a random 12-byte IV
Stores the result (salt + IV + ciphertext, all base64-encoded) in localStorage

The unencrypted key is only held in memory for the duration of a single AI request. It is never written to disk in plaintext, never logged, and never sent to Fiavaion.

When you trigger a correction, the FiavaionDictate frontend decrypts the key locally and includes it in the Authorization header of a request to /api/ai/proxy (the local Python server). The local server forwards the request to the AI provider. The key appears in memory on the local server for the duration of the request and is never persisted there.

In short: only you and the AI provider ever see your key.

Clearing your keys

To remove a single provider’s key: Settings → AI → API Keys → [Provider] → Remove

To clear all keys at once: Settings → Security → Clear All Keys

This deletes the encrypted entries from localStorage. The keys cannot be recovered after clearing.

Troubleshooting AI Setup

Ollama status badge stays grey

Make sure Ollama is actually running. Open a browser tab to http://localhost:11434 — you should see the text “Ollama is running”. If you see an error, Ollama isn’t running.
On Windows, Ollama should start automatically with Windows. Check the system tray for the Ollama icon.
On Mac, Ollama may need to be opened from your Applications folder.
If Ollama is running but FiavaionDictate still shows grey, try clicking the Refresh button in the AI settings panel.

”Model not found” error from Ollama

You have Ollama running but haven’t pulled a model yet. Run ollama list to see what you have. If the list is empty, run ollama pull gemma3:4b.

Cloud AI test fails with “Invalid API key”

Double-check that you copied the full key — they’re long and easy to truncate.
Check that you’re selecting the correct provider. A Gemini key won’t work in the Anthropic field.
For Gemini: make sure you’re using an AI Studio key (starts with AIza), not a Google Cloud service account key.
For Anthropic: make sure your account has credit added under Billing — a key from an account with no credit will return authentication errors on some models.

AI corrections seem slow

For Ollama: larger models (Mistral 7B) are slower, especially on machines without a GPU. Try Gemma 3.4B or Llama 3.2 3B for faster responses.
For cloud AI: network latency is the main factor. If you’re on a slow connection, consider Ollama for offline use.
Auto-correction triggers on each pause. If you find it interrupts your flow, switch to manual mode: Settings → AI → Correction Mode → Manual (corrections only happen when you say correct that).

”CORS error” when using cloud AI

This can happen if you’re accessing FiavaionDictate from the GitHub Pages demo URL rather than the local server. Claude and OpenAI do not support browser-direct CORS requests. Use the local server (http://localhost:8080) for those providers, or switch to Google Gemini which does support CORS.

Next Steps

Try voice commands for AI → Voice Commands Reference — AI Operations
Learn all voice commands → Full Voice Commands Reference