经 AI Skill Hub 精选评估,James-RAG-Evol 获评「推荐使用」。这款Agent工作流在功能完整性、社区活跃度和易用性方面表现出色,AI 评分 7.5 分,适合有一定技术背景的用户使用。
Local-first Graph-RAG with ontology, 3-stage security, self-evolution scaffol,提高AI工作流效率和安全性。
James-RAG-Evol 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。
Local-first Graph-RAG with ontology, 3-stage security, self-evolution scaffol,提高AI工作流效率和安全性。
James-RAG-Evol 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:pip 安装(推荐)
pip install james-rag-evol
# 方式二:虚拟环境安装(推荐生产环境)
python -m venv .venv
source .venv/bin/activate # Windows: .venv\Scripts\activate
pip install james-rag-evol
# 方式三:从源码安装(获取最新功能)
git clone https://github.com/Hashevolution/James-RAG-Evol
cd James-RAG-Evol
pip install -e .
# 验证安装
python -c "import james_rag_evol; print('安装成功')"
# 命令行使用
james-rag-evol --help
# 基本用法
james-rag-evol input_file -o output_file
# Python 代码中调用
import james_rag_evol
# 示例
result = james_rag_evol.process("input")
print(result)
# james-rag-evol 配置文件示例(config.yml) app: name: "james-rag-evol" debug: false log_level: "INFO" # 运行时指定配置文件 james-rag-evol --config config.yml # 或通过环境变量配置 export JAMES_RAG_EVOL_API_KEY="your-key" export JAMES_RAG_EVOL_OUTPUT_DIR="./output"
Local-first, auditable knowledge reasoning system with explicit reasoning paths, a sources-aware knowledge graph, and self-evolution behind a human approval gate.

한국어 README · 🚀 처음 시작하시는 분 (10살도 따라할 수 있어요)
---
JAMES combines ideas that are rarely found together:
1. Sources-aware Graph-RAG — 12 typed relations carry semantic meaning beyond embeddings, and every relation carries sources: [{doc_id, weight, role, ts}] so deleting or modifying a document surgically updates only the affected derived knowledge (Knowledge Cascade A→E, v0.3.0) 2. Cognitive Layer — cross-encoder reranker (default ON), LLM query rewriter, reflection loop (draft → critique → revise), verification engine (security + fact check), and tool router. One trace_id reconstructs the full 8-stage reasoning sequence via scripts/replay_trace.py 3. PolicyEngine as a layer, not a sprinkle — single point of role / sensitivity decisions wired into retrieval, graph, output, and tools; removing it breaks 6+ modules (v0.2 Axis 4) 4. Change Request primitive — every write (wiki edits, workspace jobs, self-evolution patches) routes through propose → review → admin approval → atomic apply → audit row. No silent writes. 5. Self-evolution behind a human gate — feedback → candidate → bench eval → human approval → deploy → auto-rollback on regression. Every deployed patch has an approver_username audit row (v0.2 Axis 5). 6. 100% local — runs on a laptop with Ollama
Each feature is regression-tested against the STEP 7 13-query baseline + RAGAS metrics. PRs touching core/{retrieval,graph,reasoning} cannot land without bench numbers.
---
| Feature | Status |
|---|---|
| Hybrid Search (Vector + BM25 + keyword + name) | Working |
| Cross-encoder reranker (MiniLM-L-6-v2) | Working — default ON (v0.3) |
| LLM query rewriter | Opt-in (v0.3) |
| Sources-aware Graph-RAG (Knowledge Cascade A→E) | Working (v0.3) |
| PolicyEngine (RBAC + ABAC + capability tokens) | Working (v0.2 Axis 4) |
| Reflection loop (draft → critique → revise) | Opt-in (v0.3) |
| Verification engine (security + fact check) | Opt-in (v0.3) |
| Tool router (read direct, write → Change Request) | Working (v0.3) |
| Change Request primitive (wiki + jobs + patches) | Working (v0.2.x + v0.3) |
| Self-evolution (human approval + auto-rollback) | Working (v0.2 Axis 5) |
Trace replay (one trace_id → full reasoning seq) | Working (v0.3) |
| Multimodal (image/video/audio + OCR-poison quarantine) | Working (v0.2 Axis 4) |
| Web search (Tavily / DuckDuckGo fallback) | Working |
| Multi-LLM routing (Ollama + Claude CLI backends) | Working |
| STEP 7 regression baseline + RAGAS | Working (v0.2 Axis 2) |
| Real-data validation (second-user gate) | Passed 2026-05-13 |
---
pip install -r requirements.txt
```bash git clone https://github.com/Hashevolution/James-RAG-Evol cd James-RAG-Evol
cp .env.example .env
James-RAG-Evol是一个开源的AI工作流,提供了Local-first Graph-RAG with ontology, 3-stage security, self-evolution scaffol功能,提高AI工作流效率和安全性。然而,项目星数较少,可能需要更多的维护和更新。
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。
AI Skill Hub 点评:James-RAG-Evol 的核心功能完整,质量良好。对于自动化工程师和运维人员来说,这是一个值得纳入个人工具库的选择。建议先在非生产环境试用,再逐步推广。
| 原始名称 | James-RAG-Evol |
| 原始描述 | 开源AI工作流:🔐 Local-first Graph-RAG with ontology, 3-stage security, self-evolution scaffol。⭐13 · Python |
| Topics | workflowai-agentfastapigraph-ragknowledge-graphlocal-aipython |
| GitHub | https://github.com/Hashevolution/James-RAG-Evol |
| License | MIT |
| 语言 | Python |
收录时间:2026-05-25 · 更新时间:2026-05-30 · 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