> ## Documentation Index
> Fetch the complete documentation index at: https://docs.vowen.ai/llms.txt
> Use this file to discover all available pages before exploring further.

# AI & API Issues

> Troubleshooting AI enhancement and API connection problems.

## API Key Not Working

### Symptoms

* "Connection failed" when verifying
* AI enhancement does nothing
* Error messages mentioning authentication

### Fix

1. Double-check your API key — copy it fresh from the provider's dashboard
2. Ensure there are no extra spaces before or after the key
3. Verify your account is active on the provider's website
4. Some providers require a payment method on file even for free tiers

## AI Enhancement Is Slow

### Symptoms

* Long delay between releasing shortcut and text appearing
* Works but takes 5+ seconds

### Possible Causes & Fixes

| Cause               | Fix                                                         |
| ------------------- | ----------------------------------------------------------- |
| Slow AI model       | Switch to a faster model (Gemini Flash Lite, Groq Llama 8B) |
| Provider overloaded | Try a different provider temporarily                        |
| Long transcription  | Shorter dictations process faster                           |
| Network latency     | Check your internet connection                              |

<Tip>For the fastest enhancement, use Groq's Llama models — they offer the lowest latency among free providers.</Tip>

## Rate Limit Errors

### Symptoms

* Works for a few transcriptions then stops
* Error mentioning "rate limit" or "quota exceeded"

### Fix

* **Wait a few minutes** — most rate limits reset within 1-5 minutes
* **Switch providers** — alternate between Groq and Gemini (both free)
* **Reduce usage** — disable AI enhancement for quick, informal dictations
* **Upgrade** — most providers offer higher limits on paid plans

<Note>If you specifically hit the **Command Mode daily cap** (5 runs/day on the Free plan), the error message says so explicitly and is different from a provider rate-limit error. [Pro](/pricing) removes the cap.</Note>

## Command Mode Returns Errors

### Symptoms

* Command Mode transcribes but doesn't produce AI output
* Error messages or garbled results

### Fix

1. Ensure your AI model has **6B+ parameters** (small models can't follow instructions)
2. Verify Command Mode is enabled in AI settings
3. Check that your shortcut isn't releasing too quickly (hold for the full duration of your command)
4. Try a known-good model like GPT-4o Mini or Gemini Flash

## Garbled Characters in Output

### Symptoms

* Characters like `?` appear in the enhanced text
* Encoding issues after AI processing

### Possible Causes

* Some third-party API proxies (like PacKYAPI) have encoding issues with streaming
* Non-UTF-8 characters in the AI response

### Fix

* Try disabling streaming (if your provider supports a non-streaming option)
* Switch to a direct provider (OpenAI, Anthropic, Groq) instead of a proxy
* Report the issue with your specific provider configuration on our Discord

## AI Not Following Instructions

### Symptoms

* Custom instructions are ignored
* AI produces unexpected formatting

### Fix

1. Keep instructions simple and specific
2. Use a more capable model (GPT-4, Claude, Gemini Pro)
3. Smaller models (7B-8B parameters) may ignore complex instructions
4. Check that your instructions aren't contradicting each other
5. Try rephrasing. Sometimes different wording works better

## Output Is Being Rewritten or Paraphrased

### Symptoms

* AI Enhancement rephrases your sentences instead of just cleaning them up
* Your wording is replaced with synonyms or restructured

### Fix

Switch to a model better at instruction following (GPT-4, Claude, Gemini Pro). Smaller open-weight models occasionally rewrite despite the system prompt forbidding it. If the issue persists, add a custom instruction like "Do not rewrite or rephrase. Only fix errors."

## AI Adds Words You Did Not Say

### Symptoms

* The enhanced output contains phrases, facts, or information you never spoke
* Hallucinated content appears in transcriptions

### Fix

Report the model and provider on our [Discord](https://discord.gg/XSGKbw4u2s). The system prompt forbids this behavior, but smaller models can still hallucinate. Switching to a larger or more capable model (GPT-4, Claude Sonnet, Gemini Pro) usually resolves it.