SuperAPI AI Directory
Partner中文AI newsFavorites

All AI tools

35,738 results · Page 500 of 993

Agent-R1

AI agents

Agent-R1 is a unified, modular agent reinforcement learning framework for training multi-step LLM agents. It trains through a step-native reinforcement learning loop: the model observes the environment, generates actions, receives tool or environment feedback, and continues executing until the task is completed or terminated. Unlike single-turn RL pipelines that treat interaction as a single ever-growing prompt-response pair, Agent-R1 models each episode as a step-level MDP transition, making tool use, environment state, context management, reward assignment, and policy optimization explicit components of the same training foundation. The framework is based on a Step-level MDP and provides three major design goals: step-level trajectory representation, flexible context management, and algorithm-system decoupling, supporting capabilities such as on-policy distillation (OPD), StepPO, multimodal vision-language model agents, and process rewards.

Read more

agent-sdk-go

AI developer platforms

agent-sdk-go is an open-source SDK for building persistent AI agents in the Go language. Its core feature is that it allows an agent's state, tool calls, and execution loop to recover after a process crash, thereby "continuing from where it left off." By default it runs in-process, providing crash recovery without additional infrastructure; when distributed, multi-process execution is needed, it can switch to Temporal or Restate. Each core component of the SDK is a pluggable interface, avoiding vendor lock-in. It supports LLM providers such as OpenAI, Anthropic, Gemini, DeepSeek, Ollama, and more; supports built-in and custom tools, MCP servers, the A2A protocol, sub-agents, human approval, multi-turn conversation history, memory and RAG, streaming output and AG-UI, extended reasoning, token usage statistics, budget control, lifecycle hooks, error control, execution configuration, distributed execution, and OpenTelemetry observability. It requires Go 1.26+ and is open-sourced under the Apache License 2.0. This project is an independent community library and is not affiliated with Temporal Technologies or Restate.

Read more

agent-skills

AI developer platforms

Agent Skills is the reference implementation of ORCA (Open Cognitive Runtime Architecture), positioned as a "runtime for executable cognition." It extracts repeatable agent reasoning logic from prompts and turns it into executable, reusable, testable, and observable Skills, allowing agents to execute directly when possible rather than relying only on prompts. The project's core concepts include: Skills (reusable cognitive workflows declared as DAGs), Capabilities (backend-agnostic contracts with typed inputs and outputs), Bindings (connecting contracts to execution backends such as PythonCall, OpenAPI, MCP, and OpenRPC), and Runtime (an execution layer with DAG scheduling, policy and security gating, CognitiveState, and traceability). Running locally requires no API key and provides a deterministic Python baseline for offline development and testing; the first run may take 30-60 seconds. The repository also provides an HTTP API, an MCP server, framework adapters (LangChain, CrewAI, AutoGen, Semantic Kernel), and integration methods such as native tool definitions for Anthropic, OpenAI, and Gemini, and is open-sourced under the Apache 2.0 license.

Read more

Agent-Wiz

AI agents

Agent-Wiz is a Python command-line tool for extracting agent workflows from mainstream AI frameworks such as LangGraph, AutoGen, and CrewAI, visualizing them, and automatically performing security threat assessments based on threat modeling methodologies such as MAESTRO. It is aimed at developers, researchers, and security teams, helping to clarify the relationships among agents, tools, and data flows in complex LLM orchestration, identify potential vulnerabilities, and generate security reports. Agent-Wiz requires an OpenAI API key to be configured in order to run analysis commands. Related framework parsers, standardized graph representations, security report generation, analysis for additional threat models such as STRIDE/PASTA/LINDDUN, and agent simulation-based threat exploration are still planned or under development.

Read more

Agent37

AI agents

Agent37 is an infrastructure platform that provides isolated sandbox hosting for AI agents. Each customer gets an independent sandbox, supports running any agent framework, is billed by the minute, and costs as little as 95% less than most sandbox providers. The platform offers persistent disks, a real Linux environment, browser and desktop access, state sleep and wake, automatic backups, usage monitoring, and spending limits, and ensures security through gVisor kernel isolation, an egress firewall, and encryption at rest. It is suitable for development teams and startups that need to build agent products for multiple tenants.

Read more

Agent4Rec

AI agents

Agent4Rec is a recommender system simulator implemented based on the paper "On Generative Agents in Recommendation" (SIGIR 2024 perspective). It contains 1,000 LLM-powered generative agents initialized from the MovieLens-1M dataset, with diverse social characteristics and preferences. Each agent interacts with personalized movie recommendations in a page-by-page manner and performs actions such as watching, rating, reviewing, exiting, and being interviewed. The simulator aims to explore the potential of LLM-empowered generative agents to simulate realistic, independent human behavior in recommendation environments, and supports multiple recommender systems (such as Random, Pop, MF, MultVAE, LightGCN) and different simulation configurations.

Read more