Agent-Graph Python SDK

The Task Manager provides Agents with the ability to create and manage scheduled tasks, enabling automated execution of workflows. It allows an Agent to act like an operations engineer by setting up unattended execution for scenarios such as scheduled reports, periodic data analysis, or regular content generation.

Key capabilities include:

• Task Specification Retrieval: Accessing guidelines for task creation, including interactive flows, JSON configuration formats, Cron expression documentation, and common use-case examples.
• Task Registration: Registering tasks from JSON documents into the scheduling system.
• Validation: The registration process validates configuration integrity, checking time formats, Cron expressions, and workflow existence.

Supported scheduling types:

• One-time Tasks (单次任务): Executes once at a specified time.
• Periodic Tasks (周期任务): Executes periodically based on a Cron expression.
• Batch Execution: Supports executing multiple workflow instances concurrently upon each task trigger, which is useful for batch processing scenarios.

The System Operations toolset provides Agents with the ability to query system resources and configurations. This allows an Agent to act like a system administrator, discovering available assets to avoid duplication and make informed decisions when designing workflows or prompts.

Key capabilities include:

• Model Discovery: List all configured model names.
• MCP Service Management: List all MCP servers, their connection status, and a summary of their tools (showing the first 3 tools for connected servers). You can also retrieve detailed configurations and full tool lists for specific servers.
• Workflow Management: List all available workflows and their descriptions, or retrieve the full configuration details (including all nodes and connection relationships) for a specific workflow.
• Prompt Management: List all prompts and their categories, or retrieve the full content of a specific prompt.
• System Tool Discovery: List the categories and functional descriptions of all available system tools.

Agent-Graph is a multi-agent platform composed of several interconnected building blocks that enable autonomous task execution, orchestration, and collaboration. The core components include:

• Agents: The autonomous execution units.
• Graphs (Workflows): The orchestration layer used to coordinate multiple agents.
• Models: The management system for LLM integrations.
• Memory: Persistent storage systems for both agent and user data (short-term and long-term).
• MCP: Integration with the Model Context Protocol to connect agents to external tools and data.
• Built-in Tools: A set of native tools for file operations, agent creation, and graph design.
• Prompt Center: A centralized repository for managing prompts used across agents and workflows.
• Conversation Features: Tools for sharing and exporting conversation data.
• Multi-User & Team: Features for team management, collaboration, and user invites.

The Graph Designer toolset enables Agents to architect complex multi-agent collaboration systems. It provides capabilities for learning design specifications, exporting existing configurations, and registering new workflows.

Key capabilities include:

• Learning Design Specifications: Agents can retrieve specifications containing node configuration parameters, connection relationships, placeholder syntax, and design principles.
• Workflow Export: Existing Graph configurations can be exported as JSON documents for inspection, modification, or optimization.
• Workflow Registration: New Graphs can be registered into the system from JSON documents. The registration process validates configuration integrity, ensuring node connections are logical and references are valid.

When designing workflows, Agents can:

• Define node roles and prompts.
• Configure models and MCP services.
• Set input/output relationships.
• Implement conditional branching using handoffs.
• Use placeholders to reference node outputs and prompt templates.

Built-in tools are categorized into three main functional groups:

1. Resource Creation Tools

Used to create and manage core platform resources.

• Agent Creator: Create and configure new Agents.
• Graph Designer: Design multi-agent workflows.
• MCP Builder: Build MCP tool servers to extend external capabilities.
• Prompt Generator: Create and manage reusable prompt templates.
• Task Manager: Set up scheduled tasks for workflow automation.

2. Collaboration Tools

Used for multi-agent interaction and file management.

• Sub-agent: Delegate tasks to specialized Agents for complex problem solving.
• File Tool: Manage files within conversations to create structured documentation.

3. Memory and Query Tools

Used for persistence and system awareness.

• Memory Tool: Store user preferences and Agent experiences to maintain continuity across conversations.
• System Operations: Query system resources and configurations (e.g., available models and tools).

The Agent Creator toolset enables Agents to design, create, and manage other Agents within the system. It allows an Agent to act as an architect, designing specialized Agents for specific tasks or optimizing existing ones by referencing design principles and system configurations.

Key capabilities include:

• Acquire Agent Creation Specifications: Retrieve guidelines that include configuration parameter descriptions, design principles, and existing system categories.
• Export Agent Configurations: Export the configuration of an existing Agent as a JSON document for inspection, analysis, or modification.
• Register New Agents: Register a new Agent into the system or update an existing Agent's configuration using a JSON document.

Through these tools, an Agent can learn to define clear capability descriptions, select appropriate models, configure MCP services and system tools, and set effective system prompts.