cloudflare/agents

Features

• Agentic AI Frameworks - An open-source framework for building stateful, durable AI agents on Cloudflare Workers with built-in storage and communication.
• Stateful Agent Runtimes - Provides a runtime for long-lived agents with persistent identity, local SQL storage, and efficient hibernation.
• Stateful Durable Objects - Provides strongly consistent, serializable state using durable objects that hibernate when idle.
• Agent State Persistence - Provides mechanisms to store and recover session data and SQL storage to ensure continuity across restarts.
• Durable Execution Runtimes - Provides a runtime environment with durable identity and local state for recoverable execution.
• Model Context Protocol - Implements the Model Context Protocol to connect AI models to local data sources and external tools.
• AI Agent State Coordination - Builds AI agents that maintain long-term memory, persistent identity, and durable sessions.
• Chat History APIs - Manages persistent conversational message logs and history to support resumable streaming sessions.
• Model Context Protocol Implementations - Implements the Model Context Protocol for standardized exchange of tools and data between agents and servers.
• Durable Multi-Step Orchestrators - Orchestrates durable multi-step pipelines that persist state and pause for human approval.
• Agent Planning and Coordination - Manages the agent planning loop, including model calls, tool selection, and decision-making flows.
• Bidirectional Agent Streams - Establishes bidirectional WebSocket communication to stream agent output and state between clients and servers.
• Strong Consistency Guarantees - Maintains strongly consistent and serializable data that persists across requests.
• Hibernation Cycles - Implements a lifecycle where agents hibernate when idle and wake on demand to support high workloads.
• Durable Task Orchestrators - Implements orchestration for multi-step tasks that persist state and support human-in-the-loop approvals.
• Stateful Agent Hosting - Provides infrastructure for hosting long-running AI agents that maintain their state across operations.
• Multi-Channel AI Deployments - Deploys a single agent across web chat, voice, email, and messaging platforms with unified state.
• Voice Agents - Manages real-time voice interactions by integrating speech-to-text and text-to-speech capabilities into the agent.
• Browser Tooling - Makes browser APIs and DOM elements available as tools to AI agents via WebSocket.
• Streaming Chat Responses - Ships a streaming chat interface that delivers AI-generated text incrementally and supports human-in-the-loop approvals.
• AI-Powered Search - Provides AI-powered search tools to retrieve relevant information and improve agent responses.
• AI Tool Monetization - Implements a dedicated framework for monetizing agent tools via the x402 micropayment protocol.
• Browser Automation Agents - Enables agents to inspect web pages and capture screenshots to perform automated actions.
• Conversational Voice Processing - Captures speech and manages real-time transcription, synthesis, and conversational interruption handling.
• Server-Side Execution Loops - Executes tool calls on the server side within a loop to maintain conversational flow.
• AI Capability Monetization - Provides a system for charging users per request for accessing specific AI agent tools and endpoints.
• Micropayment Protocols - Implements a dedicated framework for monetizing agent tools via the x402 micropayment protocol.
• Micropayment Protocol Implementations - Implements the x402 micropayment protocol to charge per-request fees for tool and endpoint invocation.
• Durable Local Storage - Runs SQL queries against durable storage for custom data retrieval and state access.
• Cron Scheduling - Provides the ability to execute background tasks on a cron schedule.
• Task Scheduling - Enables triggering compute operations at specific times or recurring intervals using programmable alarms.
• Code Execution Sandboxes - Runs generated TypeScript code inside an isolated sandbox with a virtual filesystem to prevent host side effects.
• Hibernating Connection Managers - Hibernates inactive WebSocket connections to save resources and wakes them automatically when messages arrive.
• Hierarchical Agent Routing - Implements hierarchical agent routing to distribute tasks across agent teams using nested parent-child relationships.
• Typed RPC Bridges - Provides a type-safe RPC bridge allowing clients to invoke agent logic as if it were local.
• Real-Time Synchronization - Synchronizes state and events across multiple simultaneous client connections using persistent sockets.
• Resumable Chat Sessions - Restores streaming chat sessions with persistent history after a disconnection.
• Code Sandboxing Environments - Executes generated TypeScript code within a secure, isolated sandbox featuring a virtual filesystem.
• State Change Broadcasting - Broadcasts state changes to all connected clients to maintain real-time consistency across the user interface.
• Activity Tracing - Emits structured logs and metrics to monitor and debug complex, multi-step agent activity.
• Remote Function Invocation - Exposes agent functions via a type-safe interface for remote execution across a network.
• Agent Observability - Uses distributed tracing and metrics to provide deep observability into autonomous agent decision-making and behavior.
• AI Agent Execution Monitors - Collects tracing, metrics, and structured logs to diagnose and monitor real-time agent activity.
• UI State Synchronization - Binds frontend components to agent state via hooks to ensure the UI stays current in real-time.
• WebSocket State Synchronization - Broadcasts real-time state updates via WebSockets to maintain UI consistency across sessions.

Star-Verlauf