> ## Knowledge Base Index > Fetch the complete knowledge base index at: https://help.crisp.chat/sitemap.xml > Use this file to discover available pages before exploring further. > Pure-Markdown content can be obtained by appending a '.md' suffix to the content URLs listed in the sitemap (without the trailing slash). # How to migrate from MagicReply chatbots to Hugo AI Agent *This guide explains how to migrate from legacy MagicReply chatbots to Hugo AI Agent while keeping the useful parts of your existing setup.* [Hugo AI Agent](https://crisp.chat/en/ai/) replaces the legacy MagicReply approach with a more configurable AI setup. Knowledge, behavioral instructions, routing, workflows, integrations, testing, and monitoring now have dedicated sections instead of being combined inside workflow branches and prompts. Your existing Workflows remain available while you prepare and test Hugo, so you can migrate progressively before switching your live conversations. Hugo uses AI credits, but credits are included with paid plans and you can configure workspace and per-conversation spending limits. Jump to [how to keep Hugo AI costs under control](#1-keep-hugo-ai-costs-under-control) if cost is your main concern. ||| **MagicReply migration deadline:** all workspaces still using legacy MagicReply must complete their migration to Hugo by **October 30, 2026**. **This guide is split into practical sections:** * [Why move from MagicReply to Hugo](#1-why-move-from-magicreply-to-hugo) → understand the main differences between both AI setups * [Watch the migration walkthrough](#1-watch-the-migration-walkthrough) → see the migration flow in video format * [Map your MagicReply setup to Hugo](#1-map-your-magicreply-setup-to-hugo) → understand where your existing configuration belongs * [Adapt your System Prompt and Workflows to Hugo](#1-adapt-your-system-prompt-and-workflows-to-hugo) → understand the two main changes in the migration * [Migrate your MagicReply setup](#1-migrate-your-magicreply-setup) → move each part to the appropriate Hugo feature * [Keep Hugo AI costs under control](#1-keep-hugo-ai-costs-under-control) → configure credits, Pay-As-You-Go, and spending limits * [Test Hugo before going live](#1-test-hugo-before-going-live) → validate your migrated setup safely * [Activate Hugo](#1-activate-hugo) → switch live conversations to Hugo --- # ${color}[#0080dd](Why moving from MagicReply to Hugo) MagicReply added AI-generated answers inside Workflows. Hugo turns AI into a dedicated agent that can use several configuration layers together when handling a customer conversation. The main difference is that knowledge, behavior, routing, models, integrations, and monitoring can now be configured independently. | Capability | Hugo AI Agent | MagicReply | | ---- | | **Knowledge management** | Dedicated training sources | Knowledge often included in workflows or prompts | | **Behavioral instructions** | Dedicated Instructions | Mainly configured through the System Prompt | | **Conversation routing** | Native AI Routing | Mainly handled with workflow branches and conditions | | **AI model configuration** | Dedicated model and answer settings | Limited dedicated configuration | | **External data and actions** | Integrations & MCP | Mainly workflow-based | | **Testing** | Dedicated Playground | Limited dedicated testing | | **Monitoring and debugging** | Analytics & Observability | Limited dedicated visibility | Hugo can combine these settings with the current conversation context when preparing an answer. This gives you more control over how answers are generated and lets you update one part of the AI setup without rebuilding the entire automation. For a complete overview, read [**how to navigate the Hugo AI Agent interface and features**](https://help.crisp.chat/en/article/how-to-navigate-the-hugo-interface-and-features-h1eatv/). --- # ${color}[#0080dd](Watch the migration walkthrough) This video shows the main differences between MagicReply and Hugo and how to reorganize an existing chatbot setup. ${youtube}[How to Migrate from MagicReply Chatbots to Hugo AI Agent](GUO8xtnErA4) --- # ${color}[#0080dd](Map your MagicReply setup to Hugo) MagicReply concentrated much of the AI configuration inside **Workflows** and the **System Prompt**. Hugo separates these responsibilities into dedicated areas for knowledge, behavior, routing, automation, integrations, testing, and monitoring. ![Hugo AI Agent interface sections for MagicReply migration](https://storage.crisp.chat/users/helpdesk/website/-/8/7/a/e/87ae2703583ac800/hugo-ai-agent-interface-sectio_2w5z9.png =1200xauto) __The main Hugo sections are:__ 1. **Train** → give Hugo the information it needs to answer customers 2. **Instructions** → define how Hugo should behave and communicate 3. **Routing** → decide what Hugo should do when specific situations occur 4. **Workflow Builder** → keep structured processes and deterministic automations 5. **Settings** → configure models, escalation, behavior, and conversation limits 6. **Integrations & MCP** → connect Hugo to external data and actions 7. **Playground** → test Hugo before using it with customers 8. **Analytics & Observability** → monitor performance and inspect real conversations Use this mapping when reviewing your existing MagicReply configuration: | Your MagicReply setup | Move or keep it in Hugo | | ---- | | Product information, policies, and factual answers | **Train** | | System Prompt behavioral rules | **Guidance → Instructions** | | System Prompt factual information | **Train** | | Intent-based or topic-based branches | **Guidance → Routing** | | Buttons, forms, mandatory inputs, and fixed processes | **Workflow Builder** | | AI fallback branches | **Settings and escalation** | | Account data, orders, subscriptions, or external actions | **Integrations & MCP** | | AI testing | **Evaluate → Playground** | | Monitoring and troubleshooting | **Analytics & Observability** | || You do not need to recreate your MagicReply setup identically. The migration consists of separating each responsibility and moving it to the Hugo feature designed for it. --- # ${color}[#0080dd](Adapt your System Prompt and Workflows to Hugo) The **System Prompt** and **Workflows** are the two parts of a MagicReply setup that require the biggest change in approach. #### ${color}[#445055](Move your MagicReply System Prompt) The MagicReply System Prompt could contain several types of information in the same field, including tone, behavioral instructions, product facts, business context, and rules for specific situations. With Hugo, these elements should be separated. __Move each part of your System Prompt to the appropriate Hugo section:__ * **Behavior, tone, boundaries, and response rules** → **AI Agent → Guidance → Instructions** * **Product facts, policies, documentation, and business information** → **AI Agent → Train** * **Situations that should trigger an action** → **AI Agent → Guidance → Routing** For example: `Always keep answers short and professional` belongs in **Instructions**. `Refunds are available for 30 days after purchase` belongs in **Train**. ||| Do not copy your complete MagicReply System Prompt into Hugo Instructions. Instructions should control how Hugo behaves, while factual information should be stored in training resources. Read [**how to write Hugo Instructions for AI Agent and Copilot**](https://help.crisp.chat/en/article/how-to-write-hugo-instructions-for-ai-agent-and-copilot-1qc131v/) and [**how to train Hugo AI Agent on your data**](https://help.crisp.chat/en/article/how-to-train-hugo-ai-agent-on-your-data-ijkdrl/) before moving your System Prompt. #### ${color}[#445055](Adapt your Workflows) Workflows remain available with Hugo, but they no longer need to contain the full AI conversation logic. Keep Workflows for structured processes where specific steps or actions should happen in a predictable order. __Typical Workflow use cases include:__ * Collecting required information * Showing buttons or choices * Checking conditions * Updating conversation or customer data * Triggering internal actions * Guiding users through a fixed sequence Hugo and Workflows can work together in both directions. **Hugo can start a Workflow** when a Routing rule detects a situation that requires a structured process. For example, Hugo can handle regular support questions and start a refund Workflow only when the customer asks for a refund. Read [**how to start Workflows situationally with Hugo AI Routing**](https://help.crisp.chat/en/article/how-to-start-workflows-situationally-with-hugo-ai-routing-yz715l/). **A Workflow can also hand the conversation to Hugo** with the **Answer with Hugo** block. When this block is reached, the Workflow ends and Hugo continues the conversation. Context collected during the Workflow can be passed to Hugo for the handoff. Read [**how to start Hugo from a Workflow**](https://help.crisp.chat/en/article/how-to-start-hugo-from-workflows-1i1jeof/). For the full Workflow Builder setup, read [**Getting started with Crisp Workflows**](https://help.crisp.chat/en/article/getting-started-with-workflows-hrmysb/). --- # ${color}[#0080dd](Migrate your MagicReply setup) Once you understand where each part belongs, review your current MagicReply configuration and migrate it section by section. #### ${color}[#445055](Audit your existing setup) Open your current Workflows and review your MagicReply blocks, System Prompt, branches, and fallback logic. Identify which elements are: * **Knowledge** → information Hugo needs to know * **Behavior** → rules Hugo should follow * **Decisions** → situations that should trigger an action * **Processes** → structured steps that should remain as Workflows This gives you a clear list of what needs to move and what can stay. #### ${color}[#445055](Move your knowledge to Train) Add factual information under **AI Agent → Train**. Depending on your content, you can use Questions & Answers, web pages, files, or your Crisp Knowledge Base. Avoid keeping important support information inside old Workflow branches when Hugo should be able to use it across conversations. Read [**how to train Hugo AI Agent on your data**](https://help.crisp.chat/en/article/how-to-train-hugo-ai-agent-on-your-data-ijkdrl/). #### ${color}[#445055](Move conversation decisions to Routing) Use **AI Agent → Guidance → Routing** for decisions that depend on what the customer is asking or what is happening in the conversation. A Routing rule can start a Workflow, move a conversation to another inbox, escalate the conversation, or decide when Hugo should handle a situation. Read [**how Hugo AI Agent Routing works**](https://help.crisp.chat/en/article/how-does-hugo-routing-work-47o06k/). #### ${color}[#445055](Connect live data through Integrations & MCP) Training resources are designed for information Hugo can learn and retrieve later. If an answer depends on live information such as an order, subscription, account status, or another external value, use **AI Agent → Automate → Integrations & MCP** instead. Read [**how to use Hugo Integrations**](https://help.crisp.chat/en/article/how-to-use-hugo-integrations-6z9yhh/) for supported integrations, or [**how to build MCP integrations with Hugo**](https://help.crisp.chat/en/article/how-to-build-mcp-integrations-with-hugo-tlrqmn/) to connect your own systems. Once these elements are migrated, review the System Prompt and Workflow changes described in the previous section before moving to testing. --- # ${color}[#0080dd](Keep Hugo AI costs under control) Hugo uses AI credits when the AI Agent is involved in a conversation. Monthly AI credits are already included with paid Crisp plans. You can monitor the remaining balance and usage from **AI Agent → Agent → Billing**. If the included credits are exhausted, Hugo does not create unlimited additional charges by default. Conversations can instead be escalated to human operators. You can optionally enable **Pay-As-You-Go** if you want Hugo to continue handling conversations after the included credits are exhausted. ![Hugo AI Agent billing with included credits and Pay-As-You-Go limit](https://storage.crisp.chat/users/helpdesk/website/-/8/7/a/e/87ae2703583ac800/hugo-ai-agent-billing-free-cre_s7nvr8.png =800xauto) #### ${color}[#445055](Set a Pay-As-You-Go limit) Pay-As-You-Go can have its own maximum spending limit. Go to **AI Agent → Agent → Billing** and click **Edit limit** under Pay-As-You-Go. This gives you a workspace-level ceiling for additional AI usage beyond your included credits. #### ${color}[#445055](Limit spending per conversation) You can also define how much Hugo is allowed to spend inside one conversation. Go to **AI Agent → Agent → Settings**, enable **Limit AI spend per conversation**, then enter the maximum amount. ![Set a maximum spending budget per Hugo AI Agent conversation](https://storage.crisp.chat/users/helpdesk/website/-/8/7/a/e/87ae2703583ac800/hugo-ai-agent-conversation-bud_1wzh342.png =800xauto) When the conversation budget is reached, Hugo escalates the conversation instead of continuing to consume AI credits. __You can therefore control AI spending at two levels:__ * **Pay-As-You-Go limit** → limits additional AI spending across the workspace * **Conversation budget** → limits spending inside an individual conversation || **For reference, a full conversation handled by Hugo costs around $0.10–$0.20 USD on average.** The actual amount depends on factors such as the AI model, conversation length, and complexity. Workflows themselves remain free and do not consume Hugo credits. Credits are used when Hugo or another AI-powered action is involved. For complete billing rules and usage details, read [**how Hugo AI pricing and billing work**](https://help.crisp.chat/en/article/how-does-hugo-ai-pricing-and-billing-works-h78hkv/). --- # ${color}[#0080dd](Test Hugo before going live) Once your MagicReply configuration has been migrated, go to **AI Agent → Evaluate → Playground**. The Playground lets you test the current Hugo setup without activating the AI Agent for live customer conversations. __Use real support questions and check that Hugo:__ 1. Finds the correct information 2. Follows your Instructions 3. Triggers the expected Routing rules and Workflows 4. Escalates when it should not answer 5. Handles follow-up questions correctly You can also compare AI models from the Playground to review their answers before choosing the model used by Hugo. Read [**how to choose and configure Hugo's AI model**](https://help.crisp.chat/en/article/how-to-choose-and-configure-hugos-ai-model-1m4mm57/). For conversations already handled by Hugo, use [**Hugo Observability to monitor and debug AI Agent conversations**](https://help.crisp.chat/en/article/how-to-monitor-and-debug-hugo-ai-agent-conversations-with-observability-1q79tp1/). --- # ${color}[#0080dd](Activate Hugo) When your tests are reliable, open **AI Agent → Agent → Activation**. Enable Hugo for new conversations and configure where and when the AI Agent should handle customer requests. __From the Activation settings, you can control:__ * Which conversation channels Hugo handles * The default destination for incoming conversations * How conversations are escalated to your team * When Hugo is active based on your configuration Your migrated Workflows remain available and can continue running through Hugo Routing or their other supported activation methods. For the complete activation process, read [**Getting started with Hugo AI Agent**](https://help.crisp.chat/en/article/getting-started-with-hugo-ai-agent-w6gbux/). || If Crisp support needs to inspect your migrated configuration, you can [**generate a temporary Hugo AI Agent debug link**](https://help.crisp.chat/en/article/how-to-generate-a-debug-link-for-hugo-ai-agent-csl2c4/) and share it with the support team.