Grape AI工作流 是 AI Skill Hub 本期精选Agent工作流之一。综合评分 6.8 分,整体质量稳定。我们推荐使用将其纳入你的 AI 工具库,帮助提升工作效率。
Grape AI工作流 是一套完整的 AI Agent 自动化工作流方案。通过可视化的节点编排,将复杂的多步骤任务拆解为清晰的自动化流程,实现全程无人值守的智能处理。支持与数百种外部服务和 API 无缝集成,适合构建数据处理管线、业务自动化和 AI 辅助决策系统。
Grape AI工作流 是一套完整的 AI Agent 自动化工作流方案。通过可视化的节点编排,将复杂的多步骤任务拆解为清晰的自动化流程,实现全程无人值守的智能处理。支持与数百种外部服务和 API 无缝集成,适合构建数据处理管线、业务自动化和 AI 辅助决策系统。
# 方式一:npm 全局安装 npm install -g grape # 方式二:npx 直接运行(无需安装) npx grape --help # 方式三:项目依赖安装 npm install grape # 方式四:从源码运行 git clone https://github.com/gael55x/Grape cd Grape npm install npm start
# 命令行使用
grape --help
# 基本用法
grape [options] <input>
# Node.js 代码中使用
const grape = require('grape');
const result = await grape.run(options);
console.log(result);
# grape 配置说明 # 查看配置选项 grape --config-example > config.yml # 常见配置项 # output_dir: ./output # log_level: info # workers: 4 # 环境变量(覆盖配置文件) export GRAPE_CONFIG="/path/to/config.yml"
<p align="center"> <img src="docs/assets/grape-nw.png" alt="Grape logo" width="128" /> </p>
<p align="center"> Better context transport for AI agents. </p>
<p align="center"> <a href="docs/README.md"><strong>Documentation</strong></a> · <a href="docs/v1/architecture/overview.md"><strong>Architecture</strong></a> · <a href="ROADMAP.md"><strong>Roadmap</strong></a> · <a href="CONTRIBUTING.md"><strong>Contributing</strong></a> </p>
<p align="center"> <img alt="Local-first" src="https://img.shields.io/badge/local--first-repository%20native-176f45" /> <img alt="Artifact-first" src="https://img.shields.io/badge/artifact--first-context%20packs-6f42c1" /> <img alt="CLI and MCP" src="https://img.shields.io/badge/interfaces-CLI%20%2B%20MCP-044a64" /> <img alt="Dependency-tracked" src="https://img.shields.io/badge/tracking-dependency%20hashes-111827" /> </p>
Grape is a local-first context transport layer for AI coding agents.
After MCP setup, the agent calls grape_get_context each turn with stable session identity. Grape tracks what that session has already seen, invalidates stale context when repo state changes, and ships only the safe delta (NEW, CHANGED, PINNED, RESTORE_AVAILABLE, INVALIDATE_PREVIOUS) without manual compile/diff commands. Install Grape once, configure your agent through MCP, and keep using your coding agent normally.
Instead of making agents reread the same files, rediscover the same rules, and repeat the same mistakes, Grape turns repository knowledge into dependency-tracked context artifacts that can be diffed, restored, and invalidated.
Grape is not a coding assistant, chatbot, broad agent memory platform, vector database, correctness prover, repo graph daemon, or generic search layer. It is session-scoped, proof-backed context transport: built to make coding agents cheaper to run, harder to mislead, and more consistent on real codebases.
Alpha status: the current transport slice is published as grape-context@0.1.0-alpha.3. It requires Node.js 22.13+.
For reproducible alpha.3 testing:
npm install -g grape-context@0.1.0-alpha.3
grape init --connect
The normal alpha install path is:
npm install -g grape-context
grape init --connect
grape init --connect creates .grape/, applies local SQLite migrations, captures the initial Git snapshot, reports scan diagnostics, and prints MCP integration guidance plus an agent instruction block you can paste into Cursor, Claude Code, or other MCP clients.
An MCP-capable coding agent then requests context through:
grape_get_context
Grape only omits context already sent to the same session. If the MCP client changes session ID, Grape resends rather than unsafe-omit. Restore is session-bound. Branch, source, and dependency changes may invalidate prior sent context.
For continued turns, keep the same task/query and session identity. The alpha.3 session contract is strict by design: different task wording with the same explicit session is a mismatch, and derived MCP sessions change when the query changes. See Agent Sessions for examples and recovery paths.
Manual CLI commands are debugging and fallback surfaces:
grape compile --task "Explain the files I need to edit"
grape compile --task "Explain the files I need to edit" --token-budget 4000
grape artifacts
grape artifacts --artifact <id>
grape proofs
grape proofs --proof <id>
grape claims --active
grape sessions
grape status
grape doctor
grape mcp --print-config
grape mcp --stdio
grape omitted --session <id>
grape omitted --session <id> --token <restoreToken>
grape stale
grape conflicts
grape conflicts --resolve <edge_id> --as coexists_with
grape run --session <id> -- <cmd...>
grape test --session <id> -- <cmd...>
grape bench --fixture clean-typescript-app
grape bench --fixture branch-switch-typescript-app
grape bench --fixture stale-source-typescript-app
grape bench --fixture session-reset-typescript-app
MCP exposes the same local transport path through grape mcp --stdio. Read tools include context retrieval, artifacts, claims, proofs, rules, omitted restore, stale items, conflicts, and status. Restricted write tools can record temporary candidates, command/test observations, user decisions, and confirmation requests, but they cannot promote durable truth directly.
If npm appears to keep older alpha code after installing alpha.3, clear the cache and reinstall the exact package:
npm cache clean --force
npm install -g grape-context@0.1.0-alpha.3
Grape聚焦AI代理上下文优化,方向明确。但项目初期阶段,社区活跃度低,文档完整度待提升,建议关注后��发展。
AI Skill Hub 为第三方内容聚合平台,本页面信息基于公开数据整理,不对工具功能和质量作任何法律背书。
建议在沙箱或测试环境中充分验证后,再部署至生产环境,并做好必要的安全评估。
✅ MIT 协议 — 最宽松的开源协议之一,可自由商用、修改、分发,仅需保留版权声明。
经综合评估,Grape AI工作流 在Agent工作流赛道中表现稳健,质量良好。如果你已有明确的使用需求,可以直接上手体验;如果还在评估阶段,建议对比同类工具后再做决策。
| 原始名称 | Grape |
| 原始描述 | 开源AI工作流:Better context transport for AI agents.。⭐7 · TypeScript |
| Topics | AI工作流上下文管理智能体框架 |
| GitHub | https://github.com/gael55x/Grape |
| License | MIT |
| 语言 | TypeScript |
收录时间:2026-06-13 · 更新时间:2026-06-13 · License:MIT · AI Skill Hub 不对第三方内容的准确性作法律背书。
选择 Agent 类型,复制安装指令后粘贴到对应客户端