> ## Documentation Index > Fetch the complete documentation index at: https://docs.cleve.ai/llms.txt > Use this file to discover all available pages before exploring further. # AI Chat Assistant > Context-aware AI that understands your writings and helps you create ## Overview Cleve's AI Chat is a context-aware assistant that can reference your entire knowledge base. Unlike generic AI chatbots, Cleve's AI knows your writings, ideas, and context—making it your personalized creative partner. ## Multi-Provider AI Support Choose from the best AI models based on your needs. ### Supported AI Providers | Provider | Models | Best For | | -------------- | -------------------------------- | --------------------------------------------- | | **OpenAI** | GPT-4o, GPT-4o-mini | General writing, coding, complex reasoning | | **Anthropic** | Claude 3.5 Sonnet, Claude 3 Opus | Long-form content, analysis, creative writing | | **Google** | Gemini 1.5 Pro, Gemini 1.5 Flash | Multimodal tasks, fast responses | | **DeepSeek** | DeepSeek-V3 | Coding, technical writing | | **Perplexity** | Sonar Pro | Research, fact-checking, citations | | **xAI** | Grok | Real-time information, conversational | ### Switching Models You can change AI models **mid-conversation**: 1. Click the **model dropdown** in the chat header 2. Select a different model 3. Continue the conversation seamlessly Each model maintains its own conversation history, so you can compare responses from different models. **Pro tip**: Use GPT-4o for brainstorming, Claude for long-form drafting, and Perplexity for research with citations. ## Context-Aware Responses This is where Cleve's AI shines—it can reference your actual content. ### Available Context Sources Toggle these on or off based on your needs: * **📝 Writings**: Reference your full writings for context * **💡 Ideas**: Pull from your captured ideas and notes * **🌐 Web Search**: Search the internet for real-time information * **📄 Current Document**: Use the writing you're currently editing ### How Context Works When context is enabled: 1. Your prompt is analyzed to understand intent 2. Relevant writings and ideas are retrieved using semantic search 3. The AI receives your content as context in the prompt 4. Responses are tailored to your specific knowledge base ### Example Context-Aware Prompts ``` "Summarize my writing titled 'Product Roadmap Q1'" ``` ``` "Create a LinkedIn post based on my latest article about productivity" ``` ``` "What are the common themes across all my writings about AI?" ``` ``` "Help me connect ideas from my notes on marketing and my article on storytelling" ``` Context retrieval uses semantic search with AI embeddings to find relevant content—not just keyword matching. ## Streaming Responses Watch AI responses appear in real-time as they're generated. ### Why Streaming Matters * **Faster perceived performance**: See output immediately * **Early cancellation**: Stop generation if the response goes off-track * **Better UX**: More engaging than waiting for a complete response ### Response Controls * **Stop generation**: Click the stop button to halt mid-stream * **Regenerate**: Retry the prompt with a different response * **Copy response**: One-click copy to clipboard * **Edit and retry**: Modify your prompt and regenerate ## Tools Integration The AI can take actions beyond just responding to messages. ### Available Tools * **Search Writings**: Find specific content across your knowledge base * **Update Writing**: Directly edit a writing based on instructions * **Create Writing**: Generate a new document from scratch * **Extract Info**: Pull structured data from your writings * **Text Operations**: Summarize, expand, rewrite, or translate content ### Example Tool Usage ``` User: "Update my 'Blog Ideas' writing to add a new section about AI trends" AI: I'll update that writing for you. [Uses update_writing tool] ``` ``` User: "Search my writings for mentions of 'customer feedback'" AI: Let me find those for you. [Uses search_writings tool] ``` The AI decides which tools to use automatically based on your request. ## Artifacts Panel Generated content appears in a dedicated panel for easy access. ### What Are Artifacts? When the AI creates structured content, it appears in the **Artifacts panel**: * LinkedIn posts * Twitter threads * Email drafts * Code snippets * Outlines and templates * Formatted lists ### Using Artifacts 1. AI generates content → appears in Artifacts panel 2. Review and edit the content in the panel 3. Click **Copy** to copy to clipboard 4. Click **Save as Writing** to create a new document This keeps your chat clean while giving you actionable output. Artifacts stay visible even as you scroll through chat history—perfect for referencing while you continue the conversation. ## Voice Input Speak your prompts instead of typing. ### How to Use Voice 1. Click the **microphone icon** in the chat input 2. Speak your prompt clearly 3. Cleve transcribes your speech to text using AI 4. Review and edit the transcription 5. Press Enter to send ### Voice Input Benefits * **Faster input** for long prompts * **Hands-free** when multitasking * **Natural conversation** feels more fluid * **Accessibility** for users who prefer speaking Transcription works in multiple languages and adapts to your accent over time. ## Conversation Management ### Conversation History Every chat session is automatically saved: * **Browse past conversations** from the sidebar * **Resume conversations** where you left off * **Search conversation history** by keywords * **Delete conversations** you no longer need ### Starting New Conversations Click **New Chat** to start fresh: * Previous context is cleared * Model selection resets to default * Conversation history starts clean Use this when switching topics or projects. ### Organizing Conversations * **Rename conversations** with descriptive titles * **Pin important conversations** to the top * **Archive old conversations** to reduce clutter ## Usage Tracking & Rate Limiting Different plans have different AI usage limits. ### Usage Metrics Displayed * **Messages sent this month** * **Tokens consumed** (input + output) * **Remaining quota** for your plan * **Reset date** (monthly billing cycle) ### Rate Limits by Plan | Plan | Messages/Month | Context Size | Priority | | ----------- | -------------- | ------------ | -------- | | **Free** | 50 | Limited | Standard | | **Starter** | 500 | Full | Standard | | **Pro** | 5,000 | Full | High | | **Max** | Unlimited\* | Full | Highest | \*Fair use policy applies ### When Limits Are Reached * You'll see a **usage warning** at 80% of quota * At 100%, a **paywall prompt** appears * Upgrade to a higher plan or wait for monthly reset See [Usage Limits](/pricing/overview) for details. ## Advanced Features ### Memory (Beta) Enable AI memory to have the assistant remember preferences across conversations: * **Personal details**: Writing style, tone preferences, audience * **Project context**: Ongoing work, recurring themes * **Instructions**: "Always use British spelling" or "Keep responses under 200 words" Toggle memory in chat settings. Memory persists across sessions and models. ### System Prompts Customize the AI's behavior with system-level instructions: 1. Open **Chat Settings** 2. Add a **system prompt** 3. Examples: * "You are a professional editor focused on clarity and conciseness" * "Always respond in bullet points" * "Use a friendly, conversational tone" System prompts apply to all conversations until changed. ### Temperature Control Adjust creativity vs. consistency: * **Low temperature (0.2-0.5)**: Focused, deterministic responses * **Medium (0.7)**: Balanced (default) * **High (0.9-1.0)**: Creative, varied responses Adjust in **Chat Settings → Advanced**. ## Use Cases & Examples ### Brainstorming Blog Topics ``` Prompt: "Based on my writings about productivity and AI, suggest 10 blog topics that would resonate with founders" AI: [Searches your writings, analyzes themes, suggests topics] ``` ### Drafting Social Media Content ``` Prompt: "Create a LinkedIn post based on my writing 'How I Built My First SaaS'. Make it engaging and include a hook" AI: [Generates post in Artifacts panel] ``` ### Summarizing Research ``` Prompt: "Summarize the key insights from all my writings tagged 'market research'" AI: [Retrieves relevant writings, synthesizes insights] ``` ### Expanding on Ideas ``` Prompt: "Turn my idea about 'async communication' into a full blog outline" AI: [Creates structured outline from brief idea] ``` ### Editing and Improving ``` Prompt: "Rewrite the introduction of my 'Product Launch' writing to be more compelling and add a hook" AI: [Uses update_writing tool to edit directly] ``` ## Best Practices ### Writing Effective Prompts 1. **Be specific**: "Create a 5-point outline" vs "Help me write" 2. **Provide context**: Reference specific writings or ideas 3. **Specify format**: "As a bullet list" or "In 3 paragraphs" 4. **Iterate**: Refine prompts based on initial responses ### Choosing the Right Model * **Complex reasoning**: Claude Opus, GPT-4o * **Speed**: Gemini Flash, GPT-4o-mini * **Research**: Perplexity Sonar * **Coding**: DeepSeek, GPT-4o ### Managing Token Usage * **Disable context sources** you don't need * **Use shorter prompts** when possible * **Choose smaller models** for simple tasks * **Monitor usage** to avoid hitting limits ### Context Selection * Enable **Writings** for content-heavy tasks * Enable **Ideas** for brainstorming * Enable **Web Search** for factual queries * Disable all for general conversation ## Troubleshooting ### AI Responses Are Generic **Solution**: Enable context sources (Writings, Ideas) so the AI can reference your content. ### Hitting Rate Limits Too Quickly **Solution**: * Use smaller models (mini, flash) for simple tasks * Disable unnecessary context sources * Upgrade to a higher plan ### Slow Response Times **Solution**: * Switch to faster models (Gemini Flash, GPT-4o-mini) * Reduce context size by disabling sources * Check your internet connection ### Voice Input Not Working **Solution**: * Grant microphone permissions in browser * Check browser compatibility (Chrome/Edge recommended) * Ensure microphone is not in use by another app ## Privacy & Security ### Data Handling * **Your prompts and responses** are stored encrypted in the database * **Context retrieval** happens server-side with secure queries * **AI providers** receive your prompts but not your full writings (only relevant snippets) ### Deleting Data * **Delete conversations** to remove chat history * **Disable memory** to clear AI memory * **Request data deletion** via support for full removal See our [Privacy Policy](https://cleve.ai/privacy) for details. ## Related Documentation Learn how to create and edit content that the AI can reference. Understand how semantic search powers context retrieval. Detailed comparison of all supported AI models. See rate limits and quotas for each plan.