登录

astron-agent

iflytekApache-2.0活跃维护

科大讯飞出品的商用级智能体工作流平台，用低代码方式编排多智能体与 MCP 工具链，帮企业快速搭建可落地的 SuperAgent 应用。

Enterprise-grade, commercial-friendly agentic workflow platform for building next-generation SuperAgents.

v1.1.2发布于 2026-09-07 06:43 UTC8,923star +2 / 7 天911fork同步于 2026-11-10 11:31 UTC
智能体工作流多智能体低代码编排企业级MCP
对比 GitHub 官网agentagentic-aiagentic-workflowaienterprise

怎么上手

由 README 提炼，执行前请确认
原文 
git clone https://github.com/iflytek/astron-agent.git
cd docker/astronAgent
cp .env.example .env
vim .env
docker compose -f docker-compose-with-auth.yaml up -d
硬件要求按 README 说明

README 未说明显存、CPU 或 GPU 型号等硬件要求

有 Docker 镜像
下载：源码 zip·GitHub Releases（安装包）本站不镜像安装包

项目说明

由 README 自动翻译提炼
没人工点评？写下你的经验

企业级可商用 Agentic 工作流平台，用于编排智能体、模型与工具并落地生产

科大讯飞出品的商用级智能体工作流平台，用低代码方式编排多智能体与 MCP 工具链，帮企业快速搭建可落地的 SuperAgent 应用。

核心能力

  • 编排 AI 工作流并管理模型与 MCP 工具
  • 原生集成智能 RPA，打通企业内外系统
  • 提供高可用一键部署方案，含认证网关
  • 接入讯飞开放平台海量即用工具生态
  • 支持 API 快速接入或本地 MaaS 集群部署

容易踩的坑

启动前需先完成 Casdoor 私有初始化与密码迁移，否则认证网关校验不通过；需先按文档配置 .env 环境变量；Helm 部署仍在开发中，K8s 环境暂不可用

内容由模型根据 README 生成，可能不完整或不准确，以官方文档为准。看原文 README

最近发版

5 个版本v1.1.2
全部 release
  1. v1.1.2Astron Agent v1.1.2 — Security Release最新正式版2026-09-07 · 2026-09-07 06:43 UTC

    Security Release

    Astron Agent v1.1.2 is a security release. Operators running v1.1.1 or earlier should upgrade promptly, especially those running multi-tenant deployments.

    Workflow code execution and tenant isolation

    This release addresses unsafe workflow code-node execution that could allow an authenticated user with workflow creation permissions to execute code in the workflow service container and access or modify other tenants' data in affected deployments.

    • Affected versions: <= v1.1.1, when using the legacy unsandboxed local code executor. Single-tenant deployments are also affected by the code-execution risk.
    • Patched version: v1.1.2.
    • Related advisory: GHSA-mh3w-4q3f-2fg5. The security advisory is now public. A CVE identifier has been requested from GitHub and is awaiting assignment.

    The release includes the executor migration in #1650, integrated into main through #1651, together with the preceding security hardening:

    • Reject unsafe local code execution and use the built-in isolated LangChain/Pyodide executor by default; E2B remains optional.
    • Migrate untouched legacy executor configuration while preserving explicit process-level overrides, including intentional disabling of code nodes.
    • Strengthen authentication on internal Workflow and Tenant service APIs, tenant checks, credential handling, and deployment isolation.

    Additional security fixes

    • Prevent server-side request forgery (SSRF) in outbound plugin requests (#1669).
    • Harden XSS, artifact and sandbox boundaries, and scope CSRF exclusions (#1643).

    Upgrade Notes

    1. Back up your databases, persistent data, and deployment configuration before upgrading.
    2. Update the deployment files and service images together. Use the Compose, service configuration, and gateway configuration shipped with this release, merging your site-specific settings. Updating only the source checkout or only the workflow image does not apply the complete fix.
    3. For Docker Compose deployments, set ASTRON_AGENT_VERSION=v1.1.2 in docker/astronAgent/.env, pull the updated service images, and recreate the stack. Then verify service health and workflow code-node execution. Automated vulnerability scanners may not flag this application-level issue; operators should verify the deployed image versions explicitly.
    4. Review custom code-executor overrides. CODE_EXEC_TYPE=local is no longer supported; use the default isolated executor or a supported remote executor. Code that depends on host filesystem, environment, subprocess, or network access may need adaptation to the isolated runtime.

    Container images: The v1.1.2 release workflow successfully built and pushed all eight core service images, Console Hub, and Console Frontend to GHCR with the v1.1.2 tag for linux/amd64 and linux/arm64. These images correspond to release commit 189600dd691e8562bd0b9f7112727b107b8a2e10. Use the explicit release tag for this upgrade; record or pin image digests for reproducibility.

    See the release's deployment guide and configuration reference for deployment requirements and executor migration details.

    Other Changes

    • Add Langfuse integration and end-to-end agent traces (#1607).
    • Harden dependency-aware workflow YAML imports (#1623).
    • Improve image-build reliability with a reliable PyPI source for RPA and retries for Maven Ce
  2. statistic-2026-092026-09-01 · 2026-09-01 00:32 UTC
    查看更新说明
    • payee: "@whyiug" reward: CNY RMB: 500 accounts:

      • issue: "#1575" payer: |- @FenjuFu

        currency: CNY RMB reward: 500 source: ""

  3. v1.1.1Astron Agent v1.1.12026-08-07 · 2026-08-07 09:31 UTC
    查看更新说明

    What's Changed

    Based on PRs merged into main since v1.1.0.

    New Features

    • feat(console): Integrate workflow functionality into the Agent (#1487)
    • feat(mcp): add Streamable HTTP with SSE fallback (#1552)

    Fixes

    • fix(knowledge): load platform config from persistent storage (#1558)
    • fix(workflow): allow zero LLM temperature (#1562)
    • fix(console): include space context in debug SSE requests (#1569)
    • fix(console): prevent blank workflow history poisoning (#1579)
    • fix(agent): harden CoT parsing and avoid duplicate LLM calls (#1589)

    Improvements

    • docs: add agent guidance files (#1534)
    • chore: Update README with CNCF Landscape details (#1543)
    • docs: add Japanese README (#1109)
    • docs(faq): 补齐 FAQ 缺失条目（社区问答 + 维护版 FAQ 文档同步，含截图） (#1553)
    • docs: fix translated README license links (#1546)
    • docs: register ja locale in VitePress config (#1550)
  4. v1.1.0Astron Agent v1.1.02026-07-06 · 2026-07-06 11:31 UTC
    查看更新说明

    This release expands the standalone Agent into a more capable execution surface: Agents can now use configured Workflows, Link plugins, and Mem0-backed memory directly from the standard Agent experience. It also includes security hardening for tool-debug URLs, CI/test stabilization, and documentation cleanup for the v1.1 line.

    Highlights

    Agent Capabilities

    • Workflow invocation inside Agents. Standard Agents can persist workflow capabilities, expose workflows as runtime tool callbacks, and use them from both chat and debug paths. The implementation includes ownership and space-membership validation, workflow runtime service support, response extraction handling, and test coverage for key runtime branches. (#1487)
    • Plugin capability for Standard Agents. The Agent ability page can import Link plugins from the plugin store, persist selected tools, reload them in bot details, and pass them into debug/chat requests for runtime invocation. Official/admin-owned plugins are handled as trusted import sources. (#1448)
    • Mem0 Agent memory. Added configurable Mem0 memory support for Agents, including scoped memory settings, improved retrieval, app-level memory isolation, non-blocking memory writes, Chinese memory preservation, and related CI/test fixes. (#1479)

    Security and Stability

    • Hardened official tool debug URL validation and trusted-owner handling to reduce unsafe redirect/debug invocation risk. (#1453)
    • Addressed GitHub code-scanning alert no. 54 for server-side request forgery. (#1459)
    • Stabilized Superteam CI and remote test checks, including S3 client integration-test setup and remaining remote test failures. (#1429, #1431, #1432, #1433)

    Documentation and Maintenance

    • Added the v1.0.9 release-notes PDF to homepage resources. (#1442)
    • Cleaned up unrelated documentation/files and ignored local agent instruction files in git. (#1438, #1435)

    Upgrade Notes

    • To use Agent workflow invocation, deploy the Console/Toolkit changes together so configured workflows can be resolved and executed through the new runtime path.
    • Review Agent capability settings after upgrade if you plan to enable plugins, workflow tools, or Mem0 memory for existing Agents.

    Change Scope

  5. v1.0.9Astron Agent v1.0.92026-06-18 · 2026-06-18 09:38 UTC
    查看更新说明

    This release centers on a major overhaul of the standalone Agent (agents created directly in the Console): its chat runtime was rebuilt on Spring AI and gained MCP and Skill capabilities. It also adds team publish approval, an agent workbench redesign, and a round of documentation, examples, and security improvements.

    Highlights

    Standalone Agent

    • Runtime rebuilt on Spring AI. The standalone agent's chat runtime was migrated from a hand-rolled OpenAI-compatible implementation to Spring AI (spring-ai-openai), using OpenAiChatModel and ToolCallback / ToolCallingManager for model calls, streaming, and framework-managed tool execution. Legacy runtime paths were removed. (#1418, #1425)
    • MCP (Model Context Protocol) support. Standalone agents can now connect external MCP tools by configuring MCP Server URLs on the capability page; the runtime fetches and invokes those tools and lets the model choose them by description. (#1418, #1426)
    • Skill module. A new Skill block on the capability page reuses the resource-management importable-skill list (search / multi-select / remove, up to 30). At runtime each skill exposes read_skill_* (reads SKILL.md and referenced resources) and run_skill_* (executes commands in the E2B script sandbox via a new core/agent endpoint). (#1424, #1427)

    Collaboration & Workbench

    • Added a team publish approval flow for shared spaces, with OWNER/ADMIN review of workflow/agent publishing. (#1383)
    • Redesigned the agent workbench with refined settings and interactions. (#1397)
    • Added agent debug conversation history and improved web-search tool orchestration.

    Docs, Examples & i18n

    • Standardized the VitePress docs site with an i18n layout, edit links, and a contribution guide. (#1406)
    • Added a community workflow examples gallery (scaffold, seeds, lint, docs page) and a submission issue template. (#1407, #1408)
    • Supplemented the FAQ with community-sourced Q&A (#1413) and refreshed MiniMax M-series model examples (#1373).

    Security & Stability

    • Hardened code-scanning security checks. (#1398, #1402)
    • Tightened agent tool-calling boundaries and stabilized order-dependent toolkit tests. (#1403)

    Upgrade Notes

    • Standalone agent model selection: the two built-in default Spark models were removed from agent creation — choose a model explicitly (e.g., one configured in Model Management).
    • Database migration: Flyway V1.38 adds the chat_bot_base.skills column and runs automatically on startup.
    • Skill execution (run_skill): requires an E2B script sandbox configured in resource management. The Console backend reaches core/agent via the AGENT_URL env (defaults to http://core-agent:${CORE_AGENT_PORT} in Docker). read_skill works without a sandbox.

    Change Scope

用户点评

登录用户的真实使用记录，可按版本和硬件筛选
共 0 条

    还没有人点评。第一个说出真实体验的人最有价值。

    README 原文

    未翻译
    在 GitHub 打开

    Astron Agent Logo

    Astron_Readme

    License GitHub Stars Ask DeepWiki

    English | 简体中文 | 日本語

    🔭 What is Astron Agent

    Astron Agent is an enterprise-grade, commercial-friendly Agentic Workflow development platform that integrates AI workflow orchestration, model management, AI and MCP tool integration, RPA automation, and team collaboration features. The platform supports high-availability deployment, enabling organizations to rapidly build scalable, production-ready intelligent agent applications and establish their AI foundation for the future.

    Why Choose Astron Agent?

    • Stable and Reliable: Built on the same core technology as the iFLYTEK Astron Agent Platform, providing enterprise-grade reliability with a fully available high-availability version open source.
    • Cross-System Integration: Natively integrates intelligent RPA, efficiently connecting internal and external enterprise systems, enabling seamless interaction between Agents and enterprise systems.
    • Enterprise-Grade Open Ecosystem: Deeply compatible with various industry models and tools, supporting custom extensions and flexibly adapting to diverse enterprise scenarios.
    • Business-Friendly: Released under the Apache 2.0 License, with no commercial restrictions, allowing free commercial use.

    Key Features

    • Enterprise-Grade High Availability: Full-stack capabilities for development, building, optimization, and management. Supports one-click deployment with strong reliability.
    • Intelligent RPA Integration: Enables cross-system process automation, empowering Agents with controllable execution to achieve a complete loop “from decision to action.”
    • Ready-to-Use Tool Ecosystem: Integrates massive AI capabilities and tools from the iFLYTEK Open Platform, validated by millions of developers, supporting plug-and-play integration without extra development.
    • Flexible Large Model Support: Offers diverse access methods, from rapid API-based model access and validation to one-click deployment of enterprise-level MaaS (Model as a Service) on-premises clusters, meeting needs of all scales.

    🤝 Adopters

                                         

    📰 News

    🔄 Ongoing

    📅 Past

    🚀 Quick Start

    We offer two deployment methods to meet different scenarios:

    Option 1: Docker Compose (Recommended for Quick Start)

    # Clone the repository
git clone https://github.com/iflytek/astron-agent.git

# Navigate to the Docker deployment directory
cd docker/astronAgent

# Copy environment configuration
cp .env.example .env

# Configure environment variables
vim .env

    For environment variable configuration, please refer to the documentation:DEPLOYMENT_GUIDE_WITH_AUTH.md

    # Complete the required private initialization/password migration first:
# docs/CASDOOR_PASSWORD_STORAGE.md
# Then start all services; the authentication gateway requires a passing check.
docker compose -f docker-compose-with-auth.yaml up -d
    📊 Service Access Addresses

    After startup, you can access the services at the following addresses:

    Authentication Service

    AstronAgent

    Note

    • Default Casdoor login credentials: username: admin, password: 123

    Option 2: Helm (For Kubernetes Environments)

    🚧 Note: Helm charts are currently under development. Stay tuned for updates!

    # Coming soon
# helm repo add astron-agent https://iflytek.github.io/astron-agent
# helm install astron-agent astron-agent/astron-agent

    📖 For complete deployment instructions and configuration details, see Deployment Guide

    📖 Using Astron Cloud

    Try Astron：Astron Cloud provides a ready-to-use environment for creating and managing Agents. Get quick access at https://agent.xfyun.cn.

    Using Guide：For detailed usage instructions, please refer to Quick Start Guide.

    📚 Documentation

    🤝 Contributing

    We welcome contributions of all kinds! Please see our Contributing Guide

    🌟 Star History

    📞 Support

    🌐 Related Projects

    Astron Agent is part of the iFlytek Astron open-source ecosystem. These sibling projects work well alongside it:

    • SkillHub — Self-hosted, open-source agent skill registry. Publish, version, and govern skill packages, then load and run them from Astron Agent for a managed skill lifecycle from development to production.
    • AstronRPA — Agent-ready RPA suite with out-of-the-box automation tools. Call RPA workflow nodes directly inside Astron Agent to combine intelligent agents with robotic process automation.

    📄 Open Source License

    This project is licensed under the Apache 2.0 License, allowing free use, modification, distribution, and commercial use without any restrictions.

    相关项目

    同分类里 star 最高的项目
    对比全部 →
    TradingAgents
    TauricResearch·Python

    用多个 LLM 智能体分工协作模拟金融交易决策，把研究、分析、下单等环节拆给不同角色，适合做量化交易策略的智能体实验。

    多智能体+4多智能体协作
    121776+71423kv0.7.12026-11-04 10:13 UTC
    MetaGPT
    FoundationAgents·Python

    用自然语言描述需求，就能让多个 LLM 智能体扮演产品经理、架构师、工程师等角色协作，自动产出需求文档、设计稿和代码。

    多智能体协作+4多智能体协作
    71239+349.1kv0.8.22026-01-21 10:12 UTC
    camel
    camel-ai·Python

    多智能体框架，让多个大模型智能体通过角色扮演和对话自主协作，用于研究智能体规模化规律并构建协作式 AI 系统。

    多智能体+4多智能体协作
    18024+62.1kv0.2.91a72026-11-09 10:25 UTC
    ekko-studio
    EKKOLearnAI·TypeScript

    本地优先的多智能体工作台，支持多智能体对话、编码和可视化工作流，桌面端和网页端都能用。

    多智能体+4多智能体协作
    11738+231.5kv0.8.202026-11-10 06:54 UTC