> ## Documentation Index > Fetch the complete documentation index at: https://docs.elitea.ai/llms.txt > Use this file to discover all available pages before exploring further. # Memory In **Memory**, you can configure how ELITEA manages conversation context, context editing, and automatic summarization for new conversations. To open **Memory**, in **Settings**, select **Memory**. Memory settings In **Memory**, you can see three sections: Control context limits and preserved recent messages for new conversations. Control whether ELITEA clears older tool outputs while keeping recent results. Control whether and how ELITEA summarizes older messages. ## Context Management In **Context Management**, you can set how conversation history is managed in all new conversations. Context Management Context management determines how much conversation history is retained and passed to the AI model in each request. Managing context effectively balances: * **Quality**: More context helps AI provide relevant, coherent responses * **Cost**: Token usage directly affects API costs * **Performance**: Excessive context can slow response times * **Limits**: AI models have maximum token limits (context windows) **Configuration Parameters** | Parameter | Type | Default | Range | Description | | --------------------------------------------------- | ------ | ------- | --------------- | -------------------------------------------------- | | **Enable context management for new conversations** | Toggle | ON | ON/OFF | Activates automatic context management | | **Max Context Tokens** | Number | 64000 | 1000 - 10000000 | Maximum tokens allocated for conversation history | | **Preserve Recent Messages** | Number | 5 | 1 - 99 | Minimum recent messages always retained in context | ### Enable Context Management Toggle to enable or disable automatic context management for new conversations. **When enabled:** * Context is automatically managed based on **Max Context Tokens** * System preserves recent messages as specified * Older messages are automatically summarized or removed when token limit is approached * Conversation continuity is maintained efficiently **When disabled:** * All conversation history is sent with each request (until model limit reached) * No automatic context optimization occurs * May hit model token limits in longer conversations * Higher token costs and potential performance issues Keep context management **enabled** (default) for optimal performance and cost efficiency, especially for longer conversations. ### Max Context Tokens Specifies the maximum number of tokens to use for conversation context in each AI request. * Basic unit of text that AI models process * Approximately: 1 token ≈ 4 characters or ≈ 0.75 words in English * Example: "Hello, how are you?" ≈ 5-6 tokens * Both input (context) and output (response) count toward model limits **To choose the correct value** Different AI models have different context windows: * **GPT-4.1**: 128,000 tokens * **GPT-4**: 8,192 tokens (standard) or 32,768 tokens (32k version) * **GPT-3.5 Turbo**: 16,385 tokens Set *Max Context Tokens* to **50-75% of your model's limit** to leave room for: * System instructions and prompts (your default instructions count here) * AI response generation * Safety margins to prevent hitting hard limits * **Higher values (100,000+)**: * ✔️ Better context retention for very long conversations * ✔️ AI can refer to the information from much earlier in conversation * ✘ Higher token costs per request * ✘ Slower response times * **Medium values (10,000-64,000)**: * ✔️ Good balance of quality and cost (recommended) * ✔️ Suitable for most use cases * ✔️ Efficient performance * **Lower values (1,000-10,000)**: * ✔️ Minimal token costs * ✔️ Faster responses * ✘ May lose context in longer conversations * ✘ AI may "forget" earlier discussion points * **Short Q\&A sessions**: 10,000-20,000 tokens (quick questions, brief interactions) * **Standard development**: 30,000-64,000 tokens (typical coding, documentation tasks) * **Long technical discussions**: 64,000-128,000 tokens (complex debugging, architecture reviews) * **Extensive analysis**: 100,000+ tokens (large codebase reviews, comprehensive research) ### Preserve Recent Messages Specifies the minimum number of recent conversation messages to always keep in context, regardless of token limits. 1. **Priority protection**: These messages are never removed or summarized, even if total tokens exceed Max Context Tokens 2. **Recent context**: Ensures AI always has immediate conversation context 3. **Continuity**: Maintains coherent responses even when older context is summarized 4. **Count method**: Each user message and corresponding AI response counts as 2 messages **To choose the correct value** * **Quick Q\&A (1-3 messages)**: * Each question is independent * Minimal context dependency * Lower value sufficient * **Iterative development (5-10 messages)**: * Building on previous responses * Code refinements and iterations * Medium value recommended (default 5 works well) * **Complex problem-solving (10-20 messages)**: * Multi-step troubleshooting * Extended debugging sessions * Higher value ensures continuity * Recent messages are **always** kept, even if they exceed **Max Context Tokens** * If 5 recent messages contain 70,000 tokens but Max Context Tokens = 64,000: * All 5 recent messages are still preserved * Only older messages beyond these 5 are subject to token limits * Set this value carefully to avoid unintentionally high token usage * **Start with default (5)** and monitor conversation quality * **Increase if** AI loses track of recent discussion points * **Decrease if** token costs are too high and conversations are short * **Monitor**: Check how often you need to re-explain recent context ## Context Editing In **Context Editing**, you can control whether ELITEA clears older tool outputs when the conversation context grows large, while keeping recent results. **Context Editing** is available only when **Context Management** is enabled. Context Editing ## Automatic Summarization In **Automatic Summarization**, you can configure how ELITEA performs automatic summarization for conversations. When enabled, ELITEA will automatically condense long conversations to maintain context while reducing token usage. Summarization is only active when **Context Management is enabled**. If the **Enable context management for new conversations** toggle is off, summarization settings are inactive even if the summarization toggle is on. To use summarization, make sure context management is enabled first. Automatic Summarization As conversations grow longer, they consume more tokens and approach model limits. Automatic summarization: * **Detects** when the conversation length reaches a threshold * **Generates** a concise summary of older messages * **Replaces** older messages with the summary * **Preserves** recent messages (as specified in Context Management) * **Maintains** conversation continuity while reducing tokens **Configuration Parameters** | Parameter | Type | Default | Range/Options | Description | | ---------------------------------- | -------------- | -------------------------------------------------------------------- | ------------- | ---------------------------------------------- | | **Enable automatic summarization** | Toggle | ON | ON/OFF | Activates automatic conversation summarization | | **Summarization instructions** | Multiline text | "Generate a concise summary of the following conversation messages." | - | Custom instructions for summary generation | | **Target Summary Tokens** | Number | 4,096 | 100 - 4,096 | Target token count for generated summaries | ### Enable Automatic Summarization Toggle to enable or disable automatic summarization for conversations. **When enabled:** * System monitors conversation length continuously * Automatically triggers summarization when threshold is reached * Older messages are condensed into a summary * Recent messages remain untouched * Token usage is optimized for long conversations **When disabled:** * No automatic summarization occurs * Full conversation history is maintained (subject to **Context Management** limits) * May hit token limits faster in extended conversations * Manual context management may be required Keep summarization **enabled** (default) for conversations that may become lengthy, especially for: * Extended debugging or troubleshooting sessions * Iterative development work * Multi-topic discussions * Long-running analysis or research tasks ### Summarization Instructions Custom instructions that guide how conversation summaries are generated. **What You Can Specify** * **Summary style**: Bullet points vs paragraphs, technical vs conversational * **Focus areas**: What information to prioritize in summaries * **Format requirements**: Structure, length constraints, organization * **Preservation rules**: Critical information that must never be omitted * **Context needs**: How much detail to retain for continuity To change the summarization instructions, select **Full Screen View**: Edit Summarization Instructions In the opened, editor, enter and format your instructions. To save your changes, leave the full screen view. ELITEA automatically saves your instructions as you type. **Example Custom Instructions** ``` Create a structured summary of the conversation: Format: - Use bullet points for key discussion topics - Preserve all code snippets and technical commands - Include decisions made and their rationale - List any unresolved questions or action items Content: - Focus on technical details, not social pleasantries - Maintain exact terminology and technical terms - Preserve version numbers, file paths, and configurations ``` ``` Summarize the troubleshooting conversation: 1. Problem Statement: Brief description of the original issue 2. Steps Attempted: List of solutions tried and their outcomes 3. Current Status: Where we are now in the debugging process 4. Next Actions: What to try next Keep all error messages and diagnostic output verbatim. ``` ### Target Summary Tokens Specifies the target length (in tokens) for generated conversation summaries. Here are some token references: * **256 tokens**: \~192 words - short paragraph summary * **1024 tokens**: \~768 words - moderate detail summary * **4096 tokens (default)**: \~3072 words - comprehensive summary **Original conversation:** * Messages 1-20: 45,000 tokens **After summarization:** * Summary of messages 1-15: \~4,096 tokens (Target Summary Tokens) * Original messages 16-20: 15,000 tokens (preserved recent messages) **Result:** * Total Context: \~19,096 tokens (reduced from 45,000) **To choose the correct value** **When to Use:** * Conversations are highly repetitive * Maximum token savings needed * Simple Q\&A that doesn't require much context **Effects:** * ✔️ Maximum token reduction * ✔️ Lowest summarization costs * ✔️ Fastest summary generation * ✘ May lose important details * ✘ AI may need clarification more often **When to Use:** * Balance detail retention with token savings * Standard technical discussions * General-purpose conversations **Effects:** * ✔️ Good balance of detail and conciseness * ✔️ Preserves key points and decisions * ✔️ Reasonable token costs * ✔️ Suitable for most use cases **When to Use:** * Complex technical discussions * Multi-topic conversations * Conversations with critical context that must be preserved * Detailed problem-solving sessions **Effects:** * ✔️ Maximum details are preserved (default) * ✔️ Rich context for AI to reference * ✔️ Better continuity across long conversations * ✘ Higher summarization costs * ✘ Summaries themselves consume significant tokens ## How Settings Work Together Example scenario: Long development session 1. **CONTEXT MANAGEMENT** (Efficiency): * Max Context Tokens: 64,000 * Preserve Recent Messages: 5 * Result: AI can reference extensive history (up to 64K tokens) while always keeping last 5 exchanges 2. **AUTOMATIC SUMMARIZATION** (Optimization): * Enabled: Yes * Target Tokens: 4,096 * Result: When the context approaches the token limit, older messages are automatically condensed into a 4K token summary **Combined effect**: * Conversation can continue indefinitely without hitting token limits (Context Management + Summarization) * Recent context always available for coherent responses (Preserve Recent Messages) * Token costs optimized by automatic summarization of older content ## How Settings Apply All settings automatically apply to every **new** conversation you create in Chat, Agents, and Pipelines. Settings persist across sessions and devices. **Limitations:** * **Only affects new conversations**: Existing conversations retain their original settings * **Personal settings**: Does not apply to conversations created by other users or shared conversations * **Agent/Pipeline configs**: Agent and Pipeline definitions have their own independent settings ## Best Practices **General Approach** 1. **Test Default Settings First**: * Context Management: Enabled, 64,000 tokens, 5 preserved messages * Summarization: Enabled, 4,096 target tokens 2. **Monitor Conversation Quality**: * Do you hit token limits? * Does summarization maintain enough detail? 3. **Adjust One Section at a Time**: * Adjust context limits if needed * Fine-tune summarization last 4. **Document What Works**: * Keep notes on effective configurations * Track which settings work for which types of tasks The sections interact with each other: * **Disabling Context Management** also disables **Max Context Tokens** and **Preserve Recent Messages** inputs — they are hidden * **Summarization is only active when Context Management is enabled** — if you disable context management, summarization settings are also inactive even if the toggle is on Therefore, if you want summarization, you must keep **CContext Management** enabled. **For short conversations (\< 20 messages):** * Context Management: Lower token limits (20,000-30,000) to save costs * Summarization: Can disable or leave at defaults for short conversations **For medium conversations (20-50 messages):** * Context Management: Standard limits (40,000-64,000) * Summarization: Default settings work well **For long conversations (50+ messages):** * Context Management: Higher limits (64,000-128,000) * Summarization: Critical for cost control **Quality-Focused Configuration** ``` Context Management: - Max Tokens: 100,000+ (high limit) - Preserve Messages: 10-15 (more recent context) Summarization: - Target Tokens: 4,096 (comprehensive summaries) Result: Maximum context retention, best AI responses, higher costs ``` **Cost-Focused Configuration** ``` Context Management: - Max Tokens: 20,000 (lower limit) - Preserve Messages: 3-5 (minimal recent context) Summarization: - Target Tokens: 1,024 (concise summaries) Result: Minimal token usage, lower costs, may need more clarifications ``` **Balanced Configuration (Recommended)** ``` Context Management: - Max Tokens: 64,000 (default) - Preserve Messages: 5 (default) Summarization: - Target Tokens: 4,096 (default) Result: Good quality, reasonable costs, works for most scenarios ``` ## Troubleshooting **Symptoms:** * AI behavior seems unchanged after making changes **How Saving Works:** Settings auto-save — there is no Save button to click. Each change type saves differently: * **Text fields**: save when you click/tab out of the field * **Numeric fields**: save on blur, but only if the entered value is within the valid range * **Toggles**: save immediately on toggle Watch for the **"Settings saved successfully"** toast notification to confirm each save occurred. **Diagnosis:** 1. Check whether a **"Settings saved successfully"** toast appeared after each change — if not, the save may not have triggered 2. Check whether a **"Failed to save settings"** error appeared — indicates a network or server issue 3. Verify you are testing in a **brand new** conversation — existing conversations retain their original settings 4. For numeric fields, verify the values are within the allowed ranges 5. Confirm settings are visible when reopening **Memory** after the save notification appeared **Resolution:** 1. Re-enter any field value and click outside to trigger auto-save again 2. Refresh the page and re-apply any changes that weren't confirmed with a toast 3. Create a brand new conversation (do not continue an existing one) to test the new settings 4. Clear browser cache if the page is loading stale data: Ctrl+Shift+Delete (Windows) or Cmd+Shift+Delete (Mac) **Symptoms:** * You type a new value for **Max Context Tokens**, **Preserve Recent Messages**, or **Target Summary Tokens** * No *Settings saved successfully* toast appears after clicking away * The field may show a red validation error message **Cause:** Auto-save is blocked when a field value fails validation. The form will not save until all fields contain valid values. **Valid Ranges:** | Field | Min | Max | | ------------------------ | ----- | ---------- | | Max Context Tokens | 1,000 | 10,000,000 | | Preserve Recent Messages | 1 | 99 | | Target Summary Tokens | 100 | 4,096 | **Resolution:** 1. Look for a red error message directly below the field 2. Correct the value to be within the valid range 3. Click outside the field — the save will trigger automatically once the value is valid 4. Wait for the *Settings saved successfully* confirmation toast **Symptoms:** * A red **"Failed to save settings"** toast notification appears after making a change * The change was made successfully in the UI but was not persisted **Possible Causes:** 1. Network connectivity issues 2. Session timeout (logged out) 3. Server error or maintenance 4. Browser issues or extensions blocking the request **Resolution:** 1. **Check Network**: Ensure stable internet connection 2. **Refresh Page**: Reload page and re-apply the change (Ctrl+R or Cmd+R) 3. **Check Login**: Verify you're still logged in — re-authenticate if needed 4. **Try Different Browser**: Test in incognito/private mode or different browser 5. **Clear Cache**: Clear browser cache and cookies 6. **Check Console**: Open browser developer tools (F12) → Console tab to see any error messages 7. **Contact Support**: If the issue persists, contact your administrator with error details from the console