> ## Documentation Index > Fetch the complete documentation index at: https://docs.cognigy.com/llms.txt > Use this file to discover all available pages before exploring further. # Feature parity: agents versus flow agents > Compare the features of flow agents and agents, and learn how to move a flow agent to an agent. ## Feature comparison The tables compare the features of flow agents and agents, grouped by feature area. ### Context and personalization These features control which user and conversation data an agent can use to personalize a conversation. | Feature | Flow agents | Agents | | - | - | - | | User Profiles | Contact User Profile (persistent end-user information including name, email and memories), accessible across Flows and Endpoints for personalization. | Supported | | Memory Handling | Configure which user and conversation data the agent can access: contact profile fields (**Long-Term Memory Injection**) and a Context value (**Short-Term Memory Injection**). | Partly supported. The **Contact profile information** setting on the **Guardrails** tab controls which profile fields the agent can use. There is no separate setting for Context or Input data. | | CognigyScript | JavaScript superset that gives access to Input, Context and Profile objects for dynamic conversation behavior. | Partly supported. CognigyScript works only in a custom system prompt. | ### Flow integration and handover These features connect an agent to flows and to human agents in contact centers. | Feature | Flow agents | Agents | | - | - | - | | Agents node in Flow Editor | N/A | Supported. Turn on **Enable Agents node** in Agent Settings to show the node. It ends the flow and hands the conversation to an agent. | | Run a flow as a tool | **Execute Workflow Tool** | **Execute Tool Flow** tool. End the Flow with a **Tool Result** node. | | Handover to Live Agent | The **Handover to Human Agent Tool** passes conversations to human agents in contact centers. | Not supported as a tool. Use **Route to Flow** to hand the conversation to a Flow that contains a **Handover to Human Agent** Node. | | Pre-Turn Flow | N/A | Supported. Configure it in **Hooks** on the **Advanced** tab. It runs before every agent turn and doesn't run on Voice Gateway endpoints. | ### Agent configuration These features define how you set up an agent: its persona and the model and request settings that it uses. | Feature | Flow agents | Agents | | - | - | - | | Persona | Supported | Supported | | Custom Model | Supported | Supported. Select a project language model with the **Language model** setting. | | Custom Options | JSON-based overrides for LLM request parameters not exposed in the UI (model, temperature, headers, streaming and so on). | N/A | ### Tools and knowledge These features define what an agent can do and what it knows: its tools, its knowledge retrieval, and how it selects tools. | Feature | Flow agents | Agents | | - | - | - | | Knowledge | The Knowledge Tool Node is a child node of the Flow Agent node and gives direct access to Knowledge Stores for context-aware responses. | Supported | | MCP Tool | Supported | Supported | | Source Tags & Match Type | Knowledge Source tags can dynamically filter retrieval using CognigyScript. | Not supported. | | Tool Settings - Strict Mode | When enabled, strict mode ensures that arguments passed to a tool call precisely match the expected parameters. | Not supported. | | Tool Settings - Tool Choice | Determines how the flow agent selects tools: Auto, Required or None. | N/A | ### Voice These features control how an agent handles voice conversations: speech models, interruptions, keypad input, background audio, timeouts, and tracing. | Feature | Flow agents | Agents | | - | - | - | | Voice Models (TTS/STT) | Supported | Supported | | Barge in | Supported | Supported | | DTMF | Supported | Supported | | Continuous ASR | Supported | Not supported. | | Atmosphere Sounds | Supported | Supported | | Silence Overlay | Supported | Supported | | User Input timeout | Supported | Partly supported. Use **Check in after** and **Check in up to** in voice studio. | | Flow input timeout | Supported | Supported | | Tracing | Supported | Supported only in the Interaction Panel. | ### Testing and observability These features help you test an agent and monitor its behavior: analytics, simulation, mocking, debugging, and error handling. | Feature | Flow agents | Agents | | - | - | - | | Insights | Supported | Supported. Agents don't support Goals. | | Analyser | Supported | Supported | | Simulator | Supported | Supported | | Mocking | Supported | Not Supported | | Debug Settings | Configure logging of the node configuration, knowledge results, token count, system prompt, tool definitions, and LLM latency. | Not supported. | | Error Handling | Define behavior when the flow agent fails: stop the flow, continue the flow, or go to a node, with an optional error message and logging. | Not supported. | ### Response handling These features control how the responses of an agent are stored, streamed, translated, and localized, and how the agent handles images. | Feature | Flow agents | Agents | | - | - | - | | Storage & Streaming | Configure whether LLM responses are stored in Input or Context, or streamed directly to the user using buffer flush tokens. | Not supported. The channel decides how the reply is delivered: REST endpoints return the whole reply in one response, and chat and voice channels receive it in sentence-sized chunks. | | Image Handling | Configure image understanding and how images from earlier turns are handled (Keep, Drop, Minify). | Not supported. | | Translation | Supported | Not supported. | | Localization | Supported | Not supported. | ### Lifecycle and packaging These features cover how you save, move, and duplicate agents: snapshots, packages, and cloning. | Feature | Flow agents | Agents | | - | - | - | | Snapshots | Ability to create a snapshot. | Supported | | Packages | Supported | Supported | | Clone Agent | Supported | Not supported. You can clone the tools and skills of an agent. | ## Move from a flow agent to an agent To move a flow agent to an agent, build the agent from scratch and then connect it to the parts of your solution that the flow agent used. 1. **Create an agent from scratch.** Recreate the settings of the flow agent on the agent: its persona, instructions, knowledge, tools, and guardrails. Check the [feature comparison](#feature-comparison) for features that agents don't support. For the building blocks of an agent, see [Designing agents](https://docs.cognigy.com/ai/agents/develop/designing-agents.md). 2. **Link the existing flow to the agent, if necessary.** If your solution depends on logic that you built in a flow, link the flow and the agent. To hand the conversation from the flow to the agent, add an **Agents** node at the end of the flow. To show the node, turn on **Enable Agents node** in Agent Settings. To hand it from the agent to the flow, add the **Route to Flow** tool to the agent. To show the tool, turn on **Enable Route to Flow tool** in Agent Settings. To run a flow as a tool, add the **Execute Tool Flow** tool and end the flow with a **Tool Result** node. For details, see [Bi-directional transfer](https://docs.cognigy.com/ai/agents/develop/nodes/bi-directional-transfer.md). 3. **Link the endpoint to the new agent.** Open the endpoint, go to the **Setup** tab, select the agent from the **AI Agents** group of the **Target type** list, and click **Save**. Endpoints of the MCP Server type can't target an agent. 4. **Test the agent through the endpoint.** Send a test message and check that the agent responds as expected.