agentscope-ai/agentscope

Features

• Agentic LLM Frameworks - Provides a comprehensive toolkit for building LLM-powered agents with multimodal and streaming support.
• Multi-Agent Orchestration Frameworks - Orchestrates multi-agent systems with shared memory, reasoning, and tool execution capabilities.
• Agentic Workflow Orchestration - Orchestrates complex multi-agent workflows through sequential and concurrent execution pipelines.
• Model Provider Integrations - Provides a unified interface for connecting to and orchestrating multiple language model providers for reasoning and tool use.
• Agentic Workflow Engines - Defines and executes complex agent pipelines for automated reasoning and decision-making.
• Model Provider Abstractions - Abstracts diverse language model APIs into a unified interface for text generation, reasoning, and structured output formatting.
• Multi-Agent Orchestration Systems - Provides a comprehensive framework for orchestrating autonomous agents capable of reasoning, tool execution, and memory management.
• Agent Task Routers - Direct incoming queries to specific downstream agents by forcing the routing model to produce a structured data object defining the target.
• Conversational AI Agents - Facilitates the construction of conversational agents capable of multimodal interactions, tool calling, and streaming.
• Agentic Workflow Automation - Orchestrates complex, multi-step agentic workflows through sequential and concurrent pipelines.
• Workflow Orchestrators - Structures complex multi-step tasks into sequential or concurrent execution chains to automate reasoning and decision-making processes.
• Agentic Workflow Automations - Structures complex agent behaviors through pipelines and routing to automate multi-step reasoning and decision-making.
• Message-Passing Agent Orchestrators - Uses a unified message object structure to facilitate asynchronous data exchange and state persistence between autonomous agents.
• Agent Delegation - Provides mechanisms for agents to assign sub-tasks to specialized worker agents to complete complex objectives.
• Agent Evaluation Tools - Executes systematic benchmarks against agent solutions to measure quality using custom metrics.
• Agent Memory Systems - Maintains long-term state and context retention for autonomous agents across multi-turn interactions.
• Agentic Reasoning Loops - Enables middleware-based injection into the reasoning-action loop to customize agent behavior.
• Agent Orchestrators - Manages agent lifecycles, reasoning loops, and multi-agent delegation strategies for complex workflows.
• Agent Tool Execution - Invokes defined functions to allow agents to interact with external environments, retrieve information, or perform specific actions.
• External Tool Integration - Enables agents to connect to and execute external tools and APIs to perform actions beyond native model capabilities.
• Retrieval Augmented Generation - Integrates knowledge retrieval into agent workflows to provide context-aware responses.
• Multi-Agent Orchestration Frameworks - Coordinates multiple agent instances within a shared environment to enable collaborative interactions and message broadcasting.
• RAG Frameworks - Provides base classes for implementing custom document readers and knowledge retrieval logic in RAG workflows.
• Task Planning Systems - Decomposes complex objectives into sequences of manageable sub-tasks to support systematic agent execution and planning.
• Long-term Memory Stores - Provides persistent storage mechanisms for retaining context across multiple sessions and interactions.
• Concurrent Agent Execution - Supports running multiple autonomous agents simultaneously using asynchronous processing to improve workflow efficiency.
• Execution Hooks - Injects custom logic into the agent reasoning loop to intercept, validate, or transform messages and tool calls at runtime.
• Agent State Persistence - Captures and restores the internal attributes of agents and memory components to support session continuity and recovery.
• Agent Capability Extensions - Allows injection of custom logic into agent functions to modify execution flow without altering core code.
• Agent Skill Frameworks - Manages collections of task-specific instructions injected into system prompts to enhance agent performance.
• Custom Agent Builders - Allows developers to define specialized agent behaviors by extending base classes with custom logic.
• Agent Streaming Interfaces - Converts agent-generated content into asynchronous generators for real-time processing of message chunks.
• Agent Tool Integrations - Provides a middleware layer for connecting agents to external tools via dynamic schemas.
• External Knowledge Integrators - Empowers agents to dynamically query external knowledge bases as part of their reasoning process for retrieval-augmented generation.
• Tool Calling - Uses a standardized format for models to request tool execution and return results through unified data structures.
• Structured Output Enforcements - Enforces predefined schemas on agent responses to ensure consistent, machine-readable output for programmatic use.
• Tool Schema Definitions - Extracts tool definitions and JSON schemas directly from function docstrings to enable seamless integration with language models.
• AI Agent Frameworks - Platform for building and running multi-agent systems.
• AI Agents - Framework for building multi-agent LLM applications.
• Agent State Persistence - Captures and restores runtime data snapshots for agents and memory components to support stateful recovery.
• Tool Execution Interceptors - Wraps tool calls in middleware to pre-process, validate, or transform execution inputs and outputs.
• Agent Performance Monitoring - Tracks agent execution and system interactions to optimize performance and reliability.
• Event Monitoring Streams - Exposes a unified event bus to track agent reasoning and tool usage in real-time.
• Agent-to-Agent Communication - Exchanges messages and tasks between different systems using standard communication protocols to ensure interoperability with external agents.
• Agent Memory Persistence - Maintains long-term state and continuity in agent sessions using developer-defined triggers.
• Agent Memory Storage - Supports custom memory backends by inheriting from base classes for proprietary database integration.
• Tool Access Controls - Provides granular control over tool availability at the server or function level for secure agent interactions.
• Execution Interrupts - Enables graceful cancellation of ongoing agent tasks with support for custom post-processing routines.
• Agent Routing Frameworks - Delegates incoming queries to specialized agents by wrapping them as executable tools for the routing model.
• Parallel Tool Execution - Runs multiple asynchronous tool functions concurrently to reduce total processing time for complex agent tasks.
• Chat Message Formats - Converts conversation history into the specific schema required by model providers while optionally truncating content to fit token limits.
• Code Execution Environments - Executes agent-driven code within secure, isolated environments.
• Context Injection - Attaches relevant knowledge to user prompts before processing to ensure agents have the necessary information for every interaction.
• Conversation Memory Stores - Persists message history across sessions to maintain context for multi-agent systems.
• Conversational Session Management - Isolates conversation history by tracking unique session identifiers to support concurrent interactions.
• Embedding Generators - Provides a unified interface for generating vector embeddings from text, images, or videos using multiple third-party providers.
• Human-in-the-Loop Workflows - Integrates human input into automated agent workflows to enable real-time collaboration and intervention.
• Multi-Agent Debate Frameworks - Facilitates structured debate between agents to evaluate progress and reach consensus on complex tasks.
• Conversation Formatters - Structures complex dialogue histories from multiple participants into unified message formats compatible with large language model input requirements.
• Planning - Supports the creation and management of task plans to handle interruptions and prioritize urgent requests.
• Response Streaming - Returns asynchronous generators that yield cumulative response chunks as they are produced by the model.
• Isolated Execution Sandboxes - Runs tools and code in sandboxed environments to ensure security and resource separation.
• Configuration-Driven Schemas - Parses function signatures and docstrings into structured schemas to enable dynamic tool discovery and invocation by language models.
• Monitoring and Observability - Captures telemetry data across agents, models, and tools to visualize application performance.
• Agent Execution Tracing - Provides real-time telemetry and visualization of agent reasoning and tool usage through a centralized event bus.
• Agent Optimization - Provides techniques for refining agent performance through strategic workflow configuration and optimization.
• Distributed Agent Systems - Facilitates message exchange and task coordination between distributed agents.
• External Server Connectivity - Establishes connections to remote servers using standard transport protocols to enable communication between agents and external tools.
• Task Offloaders - Delegates long-running operations to background processes, resuming agent conversations automatically upon completion.
• Model Format Converters - Provides utilities for mapping internal message structures to the specific input formats required by various language model APIs.
• Multimodal Retrieval Systems - Processes information from text and image sources to help agents answer queries based on diverse data formats.
• Prompt Formatting - Supports custom prompt formatting and message transformation strategies through extensible base classes.
• Reasoning Process Controllers - Processes internal model thinking steps alongside final text output to support complex problem-solving workflows during agent interactions.
• Tool Group Configurators - Organizes collections of related tools into groups that can be activated or deactivated to control agent capabilities.
• JSON Message Serializers - Converts message objects to and from JSON format to facilitate storage, transmission, and persistence of communication history.
• Multi-Tenant Data Management - Provides service isolation to support multiple users and concurrent sessions within a single deployment.
• Result Persistence Layers - Records performance data and execution trajectories in structured storage to track agent progress.
• Event Bus Management - Provides a centralized event bus for real-time telemetry to track agent reasoning, tool usage, and system performance.
• Execution Hooks - Provides mechanisms to register and clear custom logic hooks for controlling agent behavior across scopes.
• Tool Use and Function Calling - Retrieves and executes functions from external servers by listing capabilities and accessing callable objects for use in automated workflows.
• Tool Response Streamers - Returns tool execution results incrementally using synchronous or asynchronous generators as they are produced.
• Message Routing - Routes messages automatically between agents to eliminate manual passing within shared contexts.
• Workflow Session Persistence - Resumes sessions across workflow executions to maintain continuity for agents and components.
• Realtime Control Event Handlers - Standardizes model-specific events into consistent messages for real-time frontend integration and monitoring.
• Execution Interrupts - Provides mechanisms to intercept agent execution at runtime for user steering and custom interruption handling.
• Activity Progress Monitors - Triggers hooks during plan state changes to enable real-time tracking of task execution status.
• Observability Platform Log Exporting - Forwards execution logs to external observability platforms for centralized analysis.

Star 历史