Agents

1 Create an Agent

1.1 Manage an Agent

1. Enter the agent management interface via the "My Agent" page:

• After logging in to your AgentWeave account, navigate to the "My Agent" page. This page lists all the agents you have created.

• Select an agent: Find the agent you want to manage in the list and click it to open its details page.

• Click "Manage": On the agent's details page, click the "Manage" option. This opens the agent management interface, where you can perform various administrative actions.

2. In the management interface, you can use the left sidebar to access different management features:

• Dashboard: View the agent's performance metrics and user interactions.

• Files: Manage files and knowledge base content associated with the agent.

• History: Review the agent's activity history and user interaction records.

• Settings: Adjust the agent's configuration, such as permissions, visibility, and feature settings.

• API Key: Manage the agent's API access keys for integration and automation.

By following these steps, you can effectively manage and edit your agents, ensuring they meet your business needs and provide the best user experience.

1.2 Writing Prompts

To configure an agent that fits your needs, you'll need to write a prompt—this is the agent's "Personal & Prompt" (Personal & Prompt). The persona and prompt define the agent's core role and will consistently influence its response style in all conversations. It's recommended to clearly specify the model's role, design the response tone/style, and constrain the scope of responses so chats align with user expectations.

1. The importance of prompts

• A prompt is key to guiding an agent's behavior and output. A well-crafted prompt ensures the agent consistently maintains the intended role, style, and scope during user interactions. This not only enhances the user experience but also helps prevent irrelevant or inaccurate responses.

2. Components of a prompt

• Role/Persona Definition: Clearly define the agent's identity and responsibilities. For example, the agent could be a "Q&A expert," "creative writing assistant," or "data analyst."

• Task Instructions: Specify the concrete tasks the agent must complete, including how to handle inputs, how to reason, and how to generate outputs.

• Output Format Requirements: Define the structure and format of responses, such as whether to use bullet credits, include a summary, or apply specific markers.

• Constraints and Limitations: State the rules the agent must follow, such as using only information from search results or flagging conflicting information.

3. Tips and best practices for writing prompts

• Clarity and specificity: Avoid vague wording; use clear, specific language. For example, instead of "answer the question," say "answer the user's question using the information from the search results."

• Lead with the role: State the agent's role at the beginning of the prompt to help it adopt the correct posture quickly.

• Step-by-step guidance: Break complex tasks into simple steps in a logical order to guide the agent through the work.

• Limit the response scope: Explicitly prohibit certain content types or behaviors to further constrain responses and ensure they meet expectations.

• Iterate and optimize: Prompt writing is iterative. In practice, you may need to adjust and refine the prompt based on the agent's performance to achieve the best results.

By clearly defining roles and rules, and iterating through testing, you can tailor an AgentWeave agent's Personal & Prompt to your scenarios—keeping conversations both in character and on-target—and continually improve the user interaction experience.

1.3 Users Submitting Content to the Knowledge Base (Knowledge & Retrieval – RAG Template)

When building an agent on the AgentWeave platform, you can enable the "Allow users to submit content to the knowledge base" feature to enrich knowledge and optimize interactions. Below are the configuration and usage instructions:

1. How to enable

• Go to New Agent – Configuration and find the "Allow user submit content to knowledge base" module.

• Click the "Add" button on the right to bring up the configuration dialog (as shown in the "Allow user submit content to knowledge base" popup) and start setting up the content collection flow.

2. Configuration items explained

• Form name: Enter a concise name (e.g., "Knowledge Feedback Form," "Content Supplement Form") so users can recognize the submission scenario and clearly see that "this is the entry point for contributing content to the knowledge base."

• Collect username: When checked, users must provide a name/nickname to trace content sources and support follow-ups (e.g., "Thanks to user XXX for the industry case"). You can also leave it unchecked to collect submissions anonymously.

• Field name: Define the input fields (e.g., "Knowledge Topic," "Case Description," "References") so users know what to fill in.

• Description: Provide additional guidance (e.g., "Case description should include time, context, and outcome") to standardize formatting and improve submission quality.

• Required setting: When "Required" is checked, users must complete the field before submitting, ensuring critical information isn't missing (e.g., set "Knowledge Topic" to required to avoid directionless submissions).

• Delete item: Click the "-" on the right side of a field to remove unnecessary items, streamlining the form to fit actual needs.

3. Use cases and value

• Collaborative knowledge building: For agents focused on "industry Q&A" (e.g., legal, medical), enabling this feature lets users contribute items such as "latest regulation interpretations" or "rare case handling," continuously expanding the depth and breadth of the knowledge base and making the agent's answers more vivid and professional.

• Issue feedback: With fields like "Problem Description" and "Supplemental Suggestions," users can report gaps in the knowledge base (e.g., "topic XX not covered"), helping operators iteratively refine the knowledge system and enhance the agent's service capabilities.

By properly configuring "Allow users to submit content to the knowledge base," you can establish a collaborative "agent + users" knowledge co-creation model, allowing the knowledge base to grow dynamically with business scenarios and continuously strengthening the agent's value.

2 Conversational and Interaction (Open Chat)

1. Template selection

In the AgentWeave platform's agent templates under "Conversational & Interaction," find the "LLM-Powered Chat Builder" template. This template is designed for intelligent Q&A and conversational-interaction agents, making it ideal for rapidly building chatbots.

2. Basic information configuration

• Agent name: It's recommended to include words like "dialogue" or "chat" in the name, for example "Travel Q&A Assistant," so users can intuitively understand its function.

• Visibility: Set who can see this agent based on your needs.

• Description: Describe the conversation scenario in detail, such as "Focused on travel consulting, available 24/7 to answer itinerary and attraction questions," so users know how to use the agent.

3. Model and version selection

• The Open Chat template also involves selecting a model (e.g., OpenAI, Google) and version, with emphasis on conversational capability fit.

• For example, choosing GPT-4.5 (Preview) is suitable for handling complex conversational logic.

• Make your selection based on the desired chat experience. You can choose whether a scheduled trigger is needed.

• You can customize whether to enable conversation viewing. When enabled, you can view the conversation content generated by other users using the agent you created and published.

4. Tools and prompt configuration

• Tools: In the tools panel, if you configure tools such as get_news_summary, they can be called in real time during Open Chat. For example, when creating a "Tech News Chatbot," pair it with the get_news_summary tool so that when users ask about tech trends, the agent can query news summaries during the chat and reply.

• Prompts: Prompts should be designed around conversation, for example, "Maintain a friendly tone; when uncertain, say ‘Let me check the latest information for you' to guide tool invocation," to make the chat more natural and smooth.

3 Personalized AI (AI Personal)

1. Template selection

In the AgentWeave platform's agent templates under "Personalized AI," find the "AI Personal" template. It's suitable for building a personalized assistant or dedicated agent. You can tailor it around a user's personal needs, interests, or work tasks for productivity enhancement, habit formation, learning guidance, or life management.

2. Basic information configuration

• Agent name: It's recommended to use an actual person's name, such as "Jack," so users can intuitively understand whom they're chatting with.

• Type of AI role: Set the identity/persona of the conversational role as needed.

• Personality traits: Choose from built-in tags or define your own preferred traits, such as "cheerful, intelligent, pure."

• Interests and hobbies: Choose from built-in tags or customize preferences for the persona, such as "soccer, painting, reading."

• AI avatar: Choose different expressions for the created character. In the chat interface, the avatar on the left can change based on the conversation to reflect the AI Personal's emotions.

• You can customize whether to enable conversation viewing. When enabled, you can view the conversation content generated by other users using the agent you created and published.

3. Tools and prompt configuration

• Prompts: Design prompts around the conversational agent's personality, e.g., "Use a gentle, caring tone," to make interactions more natural and fluid.

• Voice settings: Choose from built-in voice types, or upload your own recorded voice for playback.

4 Build Agent Team

4.1 Using Workflows

On the AgentWeave platform, workflows are the core capability for connecting agents and termination components to achieve automated collaboration on complex tasks. Below is a complete guide to the process, from workflow construction and component configuration to running and debugging:

1. Workflow entry and basic form

• During the "Create Composite Agent – Configuration" step, you can start the visual workflow builder (as shown on the "Create Agent Team" configuration page). Click "Settings" to open the workflow editor. The left side shows the component library, including agents (public/self-built/shared subscriptions, etc.) and termination conditions; the right side is the canvas for dragging components and wiring logic.

2. Workflows are represented as "nodes + connections"

• Nodes: Include agents (e.g., critic_agent), tools (e.g., FunctionTool), and termination conditions (e.g., TextMentionTermination). Each node serves a specific function (agents conduct dialogue, tools execute functions, conditions determine when to terminate the flow).

• Connections: Define data flow and interaction order between nodes, determining the task execution path (e.g., the output of Agent A is passed to Agent B for processing).

• Termination condition nodes include TextMentionTermination (detect specific text to trigger termination), MaxMessageTermination (limit the maximum number of message exchanges), and OR Termination (compound evaluation of multiple conditions). Drag the relevant node onto the canvas and configure the conditions (e.g., set MaxMessageTermination to "terminate the workflow after 5 interactions") to control the execution cycle and avoid infinite loops.

3. Running and debugging workflows

• Flow validation After arranging and configuring components, click "Confirm" in the editor to save, return to the "Create Agent Team" flow, fill in basic info such as "Team Name" and "Visibility," and start a trial run. Trigger the workflow (e.g., simulate a user query) and observe:

  • Node execution order: Whether agents and tools are invoked sequentially according to the connection logic, and termination conditions are triggered as intended.

  • Data interaction: Whether parameter passing between agents and tools is accurate (e.g., whether the calculator tool correctly receives its input), and whether outputs match expectations (e.g., calculation result format, the agent's response based on the tool's output).

• If the run is abnormal, you can: Check the JSON configuration via the "View JSON" feature (see the JSON code example) and verify whether fields like provider and config are correct (e.g., model name gpt-4o-mini, tool source code syntax).

4. Typical use cases

• Multi-agent collaborative Q&A: Chain "Question Decomposition Agent" → "Domain Expert Agent" → "Conclusion Integration Agent," combined with tools (e.g., knowledge retrieval), to solve complex problems step by step (e.g., legal case analysis: break down credits of contention → call a statute library tool → integrate a defense strategy).

• Automated content generation: Agent A raises a request (e.g., "Generate a travel guide outline") → tool invocation (fetch destination data) → Agent B outputs a complete guide → termination condition evaluation (terminate if content meets criteria), creating an efficient content production pipeline.

4.2 Using the A2A Team Template

1. Function positioning A2A Team is built on a Retrieval-Augmented Generation (RAG) architecture, focusing on multi-agent collaborative retrieval and content generation. It suits scenarios like knowledge Q&A and document parsing. Through division of labor among agents, it enhances the depth and accuracy of information processing.

2. Creation process

• Select the A2A Team template agent (recommended for beginners/standardized scenarios)

  • Navigation path: Click "Functional Agents" → "+ Select" → Browse Official Library/Team Shared Agents → Check the target items → Confirm to add.

  • Steps: Click "+ Select" to open the functional agents list pop-up (e.g., Mistral, 01AI, DeepSeek, etc.). Select the functional agents to integrate (multi-select supported), then click "Confirm" to add.

• Add via input link (for custom development/external API scenarios)

  • Navigation path: Click "Add via Input" → Paste the agent URL → Configure authentication parameters → Complete verification.
  • Link format specification http://{Domain}/.well-known/agent.json

• Domain: The deployment address of the agent service (e.g., rest.agentweave.ai)

• agentId: The unique identifier of the target agent (obtain from the service provider)

• your_token (optional): If the agent requires authentication, include a token in the URL (passing it via header is recommended for better security)

4.3 Using the AutoGen Agent Template

1. Function positioning

The AutoGen Agent template is deeply aligned with the AutoGen multi-agent conversational collaboration paradigm. By flexibly orchestrating agent roles and defining dialogue triggers, it enables "autonomous multi-agent collaboration" for complex tasks (e.g., code development, multi-step decision-making). It is ideal for scenarios that need to simulate human collaborative workflows (e.g., project planning, creative brainstorming).

2. Visual orchestration operations

Component drag-and-drop and network construction: In the template configuration page, the left panel offers full agent resources (public agents: Grok, Claude, ChatGPT, etc.; private agents: those you created or that are team-shared/subscribed/favorited) and flow control components (termination conditions, branching logic). Drag agents onto the right-side canvas to build a multi-agent collaboration network (e.g., "Requirement Analysis → Solution Generation → Compliance Check").

3. When publishing an agent, ensure that any included private agents have already been published; otherwise, this agent cannot be published.

4.4 Using the AWS Step Function Template

1. Function positioning

The AWS Step Function template focuses on deep collaboration between cloud workflows and agents, bridging the AWS ecosystem (e.g., serverless tasks, AWS Lambda functions, complex process automation) with AgentWeave's agent capabilities. It is suitable for scenarios that need to integrate AWS resources (e.g., cloud data processing, AWS service orchestration plus AI decision-making).

2. Operating procedure

Click Composite Settings. Use component drag-and-drop to build the flow: the left list aggregates components such as agent resources (public, private, shared agents). Drag agents and other components onto the right-side canvas to construct an "AI decision-making + cloud workflow execution" network (e.g., "Agent analyzes AWS data → triggers a Step Functions automation task → returns results to the agent for validation").

3. When publishing an agent, ensure that any included private agents have already been published; otherwise, this agent cannot be published.

5 Space Agents

1. Template selection

In the AgentWeave platform's agent templates under "Agent Spaces," you can choose any existing space template or click to customize settings. This template is designed for multiple agents to jointly converse on a single question for intelligent Q&A and dialogue interaction. It's ideal for rapidly building chatbots that need multiple perspectives and lines of reasoning to solve problems.

2. Basic information configuration

• Agent name: It's recommended to clearly indicate the space's location or setting in the name, such as "The White House," so users intuitively understand the context in which the chat takes place.

• Topic: Set a topic based on the space type; newly added agents will converse around this topic.

• Description: Describe the conversation scenario in detail, such as "Focused on travel consulting, available 24/7 to answer itinerary and attraction questions," so users know how to use the agent.

• You can customize whether to enable conversation viewing. When enabled, you can view conversations generated by other users using the agent you created and published.

3. Manage the space

• On the details page of your created space agent, go to Manage and click Space to further enrich the space agent's content.

• Background image: Based on the custom space name, you can upload an image to use as the chat background.

• Soundscape: You can upload background music that fits the space type or your preference; mp3, m4a, and wav formats are supported, with looped playback.

• Space description: Displayed on the space agent's details page to help users clearly understand the space type and environment.

6 AgentWeave Platform Third-Party Integrations

6.1 Azure AI Foundry Agent

1. Overview

Azure AI Foundry Agent is a third-party agent tool on the AgentWeave platform, built to deeply integrate AI capabilities from Microsoft's Azure ecosystem. With it, you can quickly connect AI models deployed in Azure (such as DeepSeek - R1) to AgentWeave, breaking platform silos and enabling cross-platform AI collaboration. It seamlessly blends Azure's AI capabilities into AgentWeave's agent collaboration framework, expanding application boundaries and creating more possibilities for your business.

2. Onboarding process

• Choose the template Log in to the AgentWeave platform, go to the "Third-Party" template list, find the "Azure AI Foundry agent" card, and click "Select" to start the Azure AI agent onboarding flow.

• Configure connection details In the configuration dialog, accurately fill in the following key information:

  • URL: The access endpoint for the Azure-side AI model or service, typically the endpoint of your deployed Azure resource, in the format https://[your-resource].azurewebsites.net/api/[model-path]. Ensure network connectivity; AgentWeave will use this address to call Azure AI capabilities.

  • Model name: Defaults include pre-adapted options (e.g., "DeepSeek - R1"). Choose the model you have deployed in Azure or wish to invoke. If you use a custom model, confirm platform compatibility in advance.

  • API Key: The authentication key for the Azure resource, available in the Azure portal under the resource's "Keys" or "Access policies." AgentWeave uses this key to securely access Azure AI services.

• After completing the information, click "Next." The platform will automatically validate the configuration (including network connectivity and key validity). Upon successful validation, you will proceed to the subsequent agent integration steps.

3. Use cases and advantages

• Enterprise AI collaboration If your company has built proprietary AI models on Azure (e.g., industry-customized data analytics models), connecting them to AgentWeave via the Azure AI Foundry Agent allows business teams to invoke private AI capabilities directly within AgentWeave's familiar collaboration interface. This reduces platform switching, enhances cross-department efficiency, and ensures data security and closed-loop process management.

• Multi-cloud AI capability integration In scenarios requiring the integration of Azure, other cloud providers, and open-source AI resources, the Azure AI Foundry Agent breaks cloud platform boundaries. Within AgentWeave, it can aggregate Azure models and other third-party agents to form a "multi-cloud + multi-source" AI collaboration network. Whether for cross-cloud data-driven decision-making or complex tasks like multi-model ensemble reasoning, everything can be executed in this unified environment.

• Reuse of proprietary models Proprietary models trained on Azure (e.g., models for specific customer persona analysis) can be connected to AgentWeave via the Azure AI Foundry Agent. With AgentWeave's agent orchestration and collaborative scheduling, these models can be rapidly embedded into existing business workflows (e.g., marketing agent suites, customer service agent networks), unlocking greater value.

By leveraging the Azure AI Foundry Agent, the AgentWeave platform achieves deep integration with Microsoft's Azure AI ecosystem, delivering a new cross-cloud, dedicated, and collaborative AI application paradigm. Whether you are an enterprise IT administrator consolidating multi-cloud resources or a business team reusing proprietary models, this tool helps you unlock a more flexible and powerful AI collaboration experience. Try connecting today to bring Azure AI capabilities into your agent collaboration system and open a new chapter of intelligent collaboration.

6.2 OpenAI Agent

1. Overview

The OpenAI Agent is a key bridge for connecting the AgentWeave platform with the OpenAI ecosystem. It enables you to seamlessly integrate agents developed or managed via the OpenAI SDK (such as assistants trained with OpenAI models) into AgentWeave. Through the OpenAI Agent, you can deeply fuse OpenAI's powerful language models with AgentWeave's agent collaboration framework, allowing OpenAI-driven AI services to participate in multi-agent collaboration and empower diverse business scenarios.

2. Onboarding process

• Choose the template Log in to the AgentWeave platform, go to the "Third-Party" template list, find the "OpenAI Agent" card, and click "Select" to start the OpenAI agent onboarding process.

• Configure OpenAI connection parameters: In the configuration dialog, accurately fill in the following OpenAI-specific credentials to ensure AgentWeave can securely and correctly invoke OpenAI resources:

  • Assistant_ID: The unique identifier for the Assistant you created on the OpenAI platform. In the Assistants management module of the OpenAI console, each assistant has a unique ID used to precisely locate the OpenAI agent resource to call.

  • API Key: The API key for the OpenAI platform, which you can generate on your OpenAI account's API management page. AgentWeave uses this key to pass OpenAI authentication and initiate agent invocation requests.

• After completing the information, click "Next." The platform will automatically validate the parameters (including correctness of the OpenAI API key, existence of the Assistant_ID, and network connectivity). Once validated, you will proceed to the agent integration step.

3. Use cases and advantages

• Collaboration with proprietary models

If you have trained a proprietary language model on the OpenAI platform (e.g., a contract analysis model optimized for the legal domain), connecting it to AgentWeave via the OpenAI Agent lets business teams call that model directly within AgentWeave's collaboration environment without switching platforms. This ensures service quality in specialized scenarios and significantly improves cross-department efficiency.

• Complementary multi-model capabilities

OpenAI models excel at language understanding and generation. After integration with AgentWeave, they can collaborate with other agents (e.g., agents specialized in data analysis or image recognition). For complex tasks (such as "analyze a market report and generate a visualization plan"), the OpenAI Agent can handle text interpretation and framework generation, while other agents perform data statistics and chart creation—achieving complementary strengths and efficient task completion.

• Expansion of the agent ecosystem

With the OpenAI Agent, the AgentWeave platform further extends the boundaries of its agent ecosystem by incorporating OpenAI's rich AI resources. Whether you are an enterprise user integrating internally developed OpenAI-based agents or an individual developer exploring multi-platform AI collaboration, this tool allows you to embed OpenAI capabilities into your existing agent network and unlock more collaboration possibilities.

By leveraging the OpenAI Agent, the AgentWeave platform achieves deep integration with the OpenAI ecosystem, delivering a tailored, collaborative, and diverse AI application experience. Try connecting now to bring OpenAI-driven agents into your collaboration system and unlock greater AI value.