### Answer There are a couple of things to check when your multilingual Conversation Flow agent unexpectedly switches to English: 1. Prompt Language Instructions Setting the language to "multilingual" only affects what the agent understands (the user audio) . You need to explicitly instruct the agent in your prompt to respond in Spanish. For example, add something like

respond in Spanish

to your global prompt. [(1) ]() 2. Voice and Model Selection Ensure your selected voice supports Spanish well. Some voices perform better in certain languages. You may also want to verify your TTS model supports multilingual output properly. 3. Consider Agent Transfer for Language Handling If you're experiencing inconsistent language behavior, one approach from the community is to use agent transfers—start with a multilingual agent, and if the user speaks a specific language, transfer to a dedicated agent configured for that language with the optimal voice and model. [(2) ]() 4. Speech Normalization If you have speech normalization enabled, note that it currently supports English, Spanish, French, and German. If you selected multilingual, it will auto-detect the language and normalize accordingly. [(3) ]() Key Takeaway: The multilingual setting handles understanding the user, but you must explicitly tell the agent in your prompt what language to respond in. [(1) ]()

​ To learn more, see the sources I considered: [[1] Step 1: Configure global settings](https://docs.retellai.com/build/conversation-flow/global-setting) [[2] Discord message from @teo_bebae on 2026-02-06](https://discord.com/channels/1186424376300679209/1186462578310910112/1469268090717077608) [[3] Normalize text for speech](https://docs.retellai.com/build/normalize-text) ​

Feel free to tag @AI Evy with additional questions. I also appreciate feedback, it helps me improve.

### Answer Since you already have Spanish instructions in your global prompt but the agent still randomly switches to English, here are additional things to check: 1. Voice Selection Make sure the voice you selected supports Spanish well. Some voices perform better in certain languages—try selecting a voice specifically optimized for Spanish (e.g., a Spanish Latin America voice) . [(1) ]() 2. Node-Level Prompts In Conversation Flow, individual nodes can have their own instructions that may override or conflict with your global prompt. Check each node to ensure they also include Spanish language instructions or don't contain any conflicting English-only instructions. [(2) ]() 3. Consider a Dedicated Spanish Agent A community-recommended workaround for inconsistent multilingual behavior is to use agent transfers. You can set up a dedicated Spanish agent with a voice and model optimized for Spanish, then transfer to it when Spanish is detected. This approach has been reported to work well for maintaining language consistency. [(3) ]() 4. LLM Model Choice Some LLM models handle multilingual responses more reliably than others. You can adjust the model in your global settings—GPT-4.1 is recommended as a starting point for balanced quality and performance. [(1) ]() If the issue persists after checking these settings, I recommend posting details on the [community forum]() for the support team to troubleshoot your specific configuration.

​ To learn more, see the sources I considered: [[1] Step 1: Configure global settings](https://docs.retellai.com/build/conversation-flow/global-setting) [[2] CONVERSATION FLOW AGENT - GLOBAL PROMPT](https://discord.com/channels/1186424376300679209/1371534043157500025) [[3] Discord message from @teo_bebae on 2026-02-06](https://discord.com/channels/1186424376300679209/1186462578310910112/1469268090717077608) ​

Feel free to tag @AI Evy with additional questions. I also appreciate feedback, it helps me improve.

### Answer Yes, if you have static text in a node that's written in English, the agent will speak that exact text in English—static content doesn't get translated automatically. You have two options: Option 1: Write Static Text in Spanish If you want to use static content, you need to write it directly in Spanish. The agent will speak exactly what you provide. Option 2: Use a Prompt Instead of Static Text You can switch from static text to a prompt-based instruction. In Conversation Flow, nodes can use either static content or a prompt that lets the agent generate the response. [(1) ]() [(2) ]() For example, instead of static text like "Hello, how can I help you?", you could use a prompt instruction like: - "Greet the customer warmly and ask how you can help them. Respond in Spanish." This way, the LLM generates the response following your language instructions from the global prompt. [(2) ]() Recommendation: For consistent Spanish output, either: - Rewrite all static text in Spanish, OR - Convert static text nodes to prompt-based instructions so the agent follows your global prompt's Spanish language directive The prompt approach gives you more flexibility since the agent will follow your global language instructions automatically.

Feel free to tag @AI Evy with additional questions. I also appreciate feedback, it helps me improve.

​ To learn more, see the sources I considered: [[1] Create Conversation Flow](https://docs.retellai.com/api-references/create-conversation-flow) [[2] Retell AI’s Advanced Conversation Flow | Retell AI](https://www.retellai.com/blog/unlocking-complex-interactions-with-retell-ais-conversation-flow) ​

### Answer Here's how to use Knowledge Base in Retell AI: Step 1: Create a Knowledge Base - Navigate to your dashboard and select the "Knowledge Base" tab [(1) ]() - Click the "Add" button (or "+" icon) to create a new knowledge base [(2) ]() Step 2: Add Sources You can add three types of content: [(1) ]() - **URL**: Import content from web pages (supports single pages or entire websites) - **File**: Upload documents (PDF, TXT, DOCX, etc.) up to 50MB each - **Text**: Add custom content by pasting or typing directly Step 3: Connect to Your Agent - Open the agent editor - Locate the "Knowledge Base" section - Select the knowledge base items you want to use *Step 4: Configure Settings (Optional) * After connecting, you can adjust: - **Chunks to retrieve**: Max number of chunks (1-10, default: 3) - **Similarity Threshold**: How strict matching is (default: 0.6) For Conversation Flow Agents: You can also configure knowledge base at the node level for more specific topics—this combines with your global knowledge base. Key Point: You don't need to change your prompt—the agent automatically retrieves relevant information during calls. Limits per knowledge base: - Max 500 URLs - Max 50 text snippets - Max 25 files (50MB each) Pricing: First 10 knowledge bases are free; additional ones are $8/month each, plus $0.005 per minute of calls using knowledge base.

Feel free to tag @AI Evy with additional questions. I also appreciate feedback, it helps me improve.

​ To learn more, see the sources I considered: [[1] Knowledge Base](https://docs.retellai.com/build/knowledge-base) [[2] Simplify Support with Retell AI’s Knowledge Base | Retell AI](https://www.retellai.com/blog/knowledge-base-feature) ​

### Answer You can store three types of content in a Knowledge Base: *1. Website Content (URLs) * [(1) ]() - Single web pages or entire websites - Supports auto-refreshing (re-fetches every 24 hours) and auto-crawling for URL paths *2. Documents (Files) * - Supported formats: .bmp, .csv, .doc, .docx, .eml, .epub, .heic, .html, .jpeg, .png, .md, .msg, .odt, .org, .p7s, .pdf, .png, .ppt, .pptx, .rst, .rtf, .tiff, .txt, .tsv, .xls, .xlsx, .xml 3. Custom Text Snippets - Paste or type direct information—ideal for specific instructions or data [(2) ]() Common Use Cases for Knowledge Base Content: [(3) ]() - FAQs - Company information - Product information - Troubleshooting guides - Policy documents - Billing and account information - Release notes and updates Limits: [(1) ]() - Max 500 URLs - Max 50 text snippets - Max 25 files (each up to 50MB) - For CSV/TSV/XLS/XLSX: max 1000 rows and 50 columns Best Practice: Prefer

.md

(Markdown) files over `.txt`—well-structured Markdown is chunked and retrieved more accurately.

​ To learn more, see the sources I considered: [[1] Knowledge Base](https://docs.retellai.com/build/knowledge-base) [[2] Empower AI with Dynamic Knowledge Bases | Retell AI](https://www.retellai.com/features/knowledge-base) [[3] Simplify Support with Retell AI’s Knowledge Base | Retell AI](https://www.retellai.com/blog/knowledge-base-feature) ​

Feel free to tag @AI Evy with additional questions. I also appreciate feedback, it helps me improve.

@AI Evy can knowlage base be build with inbound call ?

### Answer No, a Knowledge Base cannot be built or populated automatically from inbound calls. Knowledge Bases are created and managed separately in the dashboard before your agent uses them. You add content to a Knowledge Base through three methods: [(1) ]() - Scrape from Webpages — Upload your website's sitemap and the system extracts content automatically - Upload Files — Import documents (PDFs, Word docs, spreadsheets, etc.) - Manual Input — Copy and paste text for custom entries The Knowledge Base is a pre-configured repository that your agent retrieves information from during calls—it doesn't collect or store information from conversations. If you want to capture information from inbound calls (like customer details or conversation data) , you would use: - Dynamic Variables to extract and store call-specific data - Custom Functions to send data to external systems (CRMs, databases, etc.) [(2) ]() - Post-Call Analysis to extract insights after calls end [(3) ]()

​ To learn more, see the sources I considered: [[1] Simplify Support with Retell AI’s Knowledge Base | Retell AI](https://www.retellai.com/blog/knowledge-base-feature) [[2] Integrate Function Calling](https://docs.retellai.com/integrate-llm/integrate-function-calling) [[3] Terminologies](https://docs.retellai.com/ai-qa/terminologies) ​

Feel free to tag @AI Evy with additional questions. I also appreciate feedback, it helps me improve.

### Answer No, a Knowledge Base is not the right place for agent behavior instructions like how to capture emails, when to offer follow-ups, or conditional logic. What Knowledge Base is for: Knowledge Base stores information that your agent retrieves to answer questions—like FAQs, product details, troubleshooting guides, and policy documents. [(1) ]() [(2) ]() It provides context for responses, not behavioral instructions. Where to put agent instructions: For the behaviors you described, use these features instead: *1. Prompts (Global or Node-Level) * Instructions like "capture the customer's email" or "offer a follow-up if the customer expresses interest" belong in your agent's prompt—either the global prompt or node-level prompts in Conversation Flow. 2. Dynamic Variables Use dynamic variables to capture and store information like emails and customer names during the call. You can reference them later using

{{customer_name}}

or

{{email}}

syntax. [(3) ]() 3. Extract Dynamic Variables The Extract Dynamic Variable feature lets you automatically save names, emails, and other data mid-conversation. Variable types include text, numbers, boolean, and enum. [(4) ]() 4. Custom Functions For conditional follow-ups or sending captured data to external systems (CRMs, etc.) , use custom functions to call your APIs with the collected information. [(5) ]() 5. Conversation Flow Transitions For conditional logic ("if X, then do Y") , use equation-based transitions in Conversation Flow to route the conversation based on extracted variables. [(4) ]() Summary: K

nowledge Base = information retrieval. Agent behavior and data capture = prompts, dynamic variables, and custom functions.

​ To learn more, see the sources I considered: [[1] Knowledge Base](https://docs.retellai.com/build/knowledge-base) [[2] Simplify Support with Retell AI’s Knowledge Base | Retell AI](https://www.retellai.com/blog/knowledge-base-feature) [[3] Dynamic Variables](https://docs.retellai.com/build/dynamic-variables) [[4] Platform Changelogs | Retell AI](https://www.retellai.com/changelog) [[5] Custom Function](https://docs.retellai.com/build/conversation-flow/custom-function) ​

Feel free to tag @AI Evy with additional questions. I also appreciate feedback, it helps me improve.

### Answer Knowledge Base has its own separate pricing structure: Knowledge Base Pricing: [(1) ]() - $0.005 per minute of knowledge base usage (charged when your agent uses the KB during calls) - First 10 knowledge bases are free per workspace - $8/month for each additional knowledge base beyond the free allocation Regarding the LLM Token Surcharge: The LLM token surcharge (Rule 2) applies when your agent's prompt exceeds 3,500 tokens. [(2) ]() The token calculation includes: - Global prompt - Functions (tool descriptions) - State/node prompt - Transcript between agent and user - Tool call history and results The Knowledge Base content itself is retrieved dynamically during calls—it's not loaded entirely into your prompt. The agent retrieves relevant "chunks" based on similarity matching, so even a large Knowledge Base doesn't mean all of it counts toward your prompt tokens. However, the retrieved chunks that get added to the context during a call would contribute to the overall token count. If your combined prompt (including retrieved KB content) exceeds 3,500 tokens, the scaling factor would apply. In summary: Knowledge Base has a flat $0.005/min usage fee. The LLM token surcharge could apply if retrieved KB content plus your other prompt elements exceed 3,500 tokens total.

Feel free to tag @AI Evy with additional questions. I also appreciate feedback, it helps me improve.

​ To learn more, see the sources I considered: [[1] Simplify Support with Retell AI’s Knowledge Base | Retell AI](https://www.retellai.com/blog/knowledge-base-feature) [[2] Exceptions to Our Per-Minute Pricing](https://docs.retellai.com/accounts/billing-exceptions) ​

​ To learn more, see the sources I considered: [[1] Function Node Overview](https://docs.retellai.com/build/conversation-flow/function-node) [[2] End call](https://docs.retellai.com/build/single-multi-prompt/end-call) ​

### Answer Yes, you can use the End Call function in Conversation Flow, but it works differently than in single/multi-prompt agents. How End Call Works in Conversation Flow: In Conversation Flow, you don't trigger functions directly from the global prompt. Instead, you use a Function Node to call the End Call function. [(1) ]() The Function Node is specifically designed to call functions (both pre-built like End Call and custom functions) . When the conversation reaches that node, the associated function gets executed. To set up End Call in Conversation Flow: 1. Add the End Call tool to your agent's functions section [(2) ]() 2. Create a Function Node in your flow and associate it with the End Call function [(1) ]() 3. Set up transitions from your conversation nodes to route to this Function Node when termination conditions are met Key difference from single/multi-prompt agents: In single/multi-prompt agents, you can instruct the LLM in the prompt to call

end_call

when certain conditions are met (like when the user says "goodbye") . [(2) ]() In Conversation Flow, you design the flow so that the conversation transitions to a Function Node containing the End Call function based on your transition conditions. Tip: You can write transition conditions based on conversation context to determine when to route to your End Call node.

Feel free to tag @AI Evy with additional questions. I also appreciate feedback, it helps me improve.