ResolveAgent
2026 Agent Engineering | 面向问题解决的生产级 AIOps 智能体平台
Production-grade AIOps Mega-Agent Platform for Troubleshooting & Resolution
🧠 Intelligent Selector · 🌳 Hybrid Planner · 💾 Hierarchical Memory · 🔍 FTA Engine · 🔧 ToolHub · 🔄 Loop Engineering
✨ At a Glance | 项目速览
| 维度 | 数据 |
|---|---|
| 技术栈 | Go 1.25 平台服务 + Python 3.11+ Agent 运行时 + React WebUI(23 页)+ 移动端示例 |
| 分析路由 | 4 条业务路径:FTA 故障树推理 / Skill 技能编排 / RAG 检索增强 / 代码分析(+ LLM 直答兜底) |
| FTA 引擎 | 5 种门类型(AND/OR/VOTING/INHIBIT/PRIORITY_AND 时序语义)+ 最小割集 + 蒙特卡洛仿真 |
| 记忆体系 | 3 层架构:Working(进程内)/ Episodic(Redis TTL)/ Long-term(Milvus/Qdrant LRU) |
| 数据层 | PostgreSQL 16 表 6 组 / Redis / NATS / Milvus-Qdrant 双后端向量库 |
| 质量工程 | 480+ Python 单元测试项 + 8 套集成测试 + Go -race 测试 + Go E2E + OpenAPI 契约测试 |
📖 代码级架构讲解见 Architecture Deep Dive,功能就绪状态见 Feature Status。
What is ResolveAgent?
ResolveAgent is a production-grade AIOps Mega-Agent platform built on 2026 Agent Engineering best practices. It automates the entire troubleshooting and resolution lifecycle: from ticket ingestion and intent routing, through knowledge retrieval and fault-tree analysis (FTA), to remediation execution and feedback closure.
ResolveAgent is designed for teams running complex distributed systems who need to compress the time from alert → root cause → fix from hours to minutes, while continuously capturing tribal knowledge into reusable skills and RAG corpora.
核心闭环 | The Resolution Flywheel
ResolveAgent 的独特机制不是"单路径 AI 问答",而是一个自增强的排查知识飞轮:
告警 / 工单
│
▼
意图识别 ── Intelligent Selector:意图分析 → 上下文增强
│
▼
路由分发 ── FTA / Skill / RAG / 代码分析 / LLM 直答
│
▼
多路求解 ──▶ 根因结论 + 修复建议 + 证据链
│
▼
语料写回 ── RAG 文档 · 技能 · 长期记忆
│
└──▶ 权重自适应,回流至下一次「路由分发」
用得越多 → 知识越厚 → 排查越准
每次排查的结果都会回流为语料(RAG 文档)、技能(Skill)与长期记忆,并驱动路由权重自适应调整——平台随使用变得更聪明。
Why ResolveAgent?
Traditional AI chatbots give a single answer path. ResolveAgent introduces multi-path intelligent routing combined with a self-reinforcing knowledge loop:
| Differentiator | What it means for you |
|---|---|
| Multi-path routing | Every request is analyzed and routed to the best engine: FTA, skill execution, RAG, code analysis, or direct LLM reasoning. |
| Resilient routing | Failures are not dead ends; the selector learns from them, retries with enriched context, and adapts route weights over time. |
| Formal fault-tree reasoning | FTA engine with five gate types (AND/OR/VOTING/INHIBIT/PRIORITY_AND with dynamic ordering semantics), minimal cut sets, and Monte-Carlo simulation with Wilson confidence intervals. |
| Knowledge self-reinforcement | Every resolution enriches skills, RAG documents, and memory, so the platform gets smarter with use. |
| Full-chain auditability | Intent classification, route decisions, evidence solving, and corpus write-back are all logged with context snapshots — replayable and reviewable. |
| Production-grade resilience | Circuit breakers, fallback cascades, structured observability, and OpenTelemetry tracing out of the box. |
| Polyglot runtime | Go platform services for scale, Python runtime for AI/ML, and React WebUI for operators. |
How It Works: 一条告警的旅程
以一次典型的 K8s Pod 告警为例,ResolveAgent 端到端的处理链路:
① 接入 ② 意图分析 ③ 路由决策 ④ 多路求解
┌─────────┐ ┌──────────────┐ ┌─────────────────┐ ┌──────────────────────┐
│ 告警/工单 │─▶│ Intelligent │─▶│ Route Decision │─▶│ FTA: 故障树建模 │
│ 文本/事件 │ │ Selector │ │ │ │ Skill: k8s-pod-crash │
└─────────┘ │ · 意图分类 │ │ fta / skill / │ │ RAG: 运行手册检索 │
│ · 上下文增强 │ │ knowledge / │ │ Code: 调用链分析 │
│ (记忆/偏好/ │ │ code_analysis / │ │ Reasoning: LLM 直答 │
│ 代码检测) │ │ reasoning │ └──────────┬───────────┘
└──────────────┘ └─────────────────┘ │
▼
⑦ 权重自适应 ⑥ 语料写回 ⑤ 证据结论
┌────────────┐ ┌─────────────────────┐ ┌─────────────────────┐
│ Adaptive │◀─│ · RAG 文档沉淀 │◀───────│ 根因 + 最小割集 + │
│ Weight │ │ · 技能沉淀 │ │ 修复建议 + 证据链 │
│ Adjuster │ │ · 长期记忆沉淀 │ └─────────────────────┘
└────────────┘ └─────────────────────┘
关键设计:任何一路失败都不是终点。ResilientSelector 会记录失败上下文,按错误分类切换偏好路径(如 RAG 资源缺失 → 偏好 LLM 推理兜底),并排除已尝试路径重试,最多 3 次;全程决策留痕于审计日志,可回放、可复盘。
Use Cases
ResolveAgent is built for SREs, platform engineers, and operations teams who need reliable, evidence-based incident resolution:
- Kubernetes incident response — Pod crashes, OOMKilled, CrashLoopBackOff, network partitions.
- API / service degradation — 5xx spikes, latency regressions, dependency failures.
- Root-cause analysis (RCA) — Fault-tree modeling with minimal cut sets and probabilistic simulation.
- Knowledge-base Q&A — RAG over runbooks, post-mortems, codebases, and call-chain corpora.
- Code-level diagnosis — Static analysis, call-graph traversal, and solution document generation.
- Ticket triage and summarization — Automatic classification, routing, and structured summarization.
以下场景全部基于已接线可用的能力(真实端点/页面),可按 Quick Start 启动后直接复现:
Case 1 · K8s 告警值班响应(SRE) — 告警触发后,值班工程师在 WebUI Playground(或 POST /v1/agents/{id}/execute)提交事件描述。MegaAgent 经智能选择器自动路由:命中场景技能时,排查引擎按 manifest 定义的步骤流执行——命令步骤在沙箱中真实运行(如 kubectl get pods),stdout/stderr 原样进入诊断证据;最终产出四要素结构化解法(症状 / 关键信息 / 排查步骤 / 解决步骤),一键沉淀到 Solutions 知识库,下次同类告警直接语义检索复用。
Case 2 · 故障根因量化分析(稳定性团队) — 在 FTAEngine 页面构建故障树(五种门类型,含时序与优先级语义),引擎计算最小割集定位最脆弱路径,蒙特卡洛仿真给出顶事件失效概率与 Wilson 置信区间。分析结果持久化到 /api/v1/fta/documents/{id}/results,作为架构评审与容量决策的量化依据。
Case 3 · 运维知识库问答(技术支持 / 新人上手) — 用 POST /v1/corpus/import 把 runbook、复盘报告、代码仓库批量导入 RAG(Milvus/Qdrant 向量库,自动分块入库);团队在 RAG 页面或 POST /v1/rag/query 自然语言提问,检索命中带出处引用。排查产出的解法经 /v1/solutions/sync-rag 回写知识库,形成「用得越多、答得越准」的语料飞轮。
Case 4 · 代码级诊断(研发团队) — POST /v1/code-analysis/static 对目标服务做静态分析与错误诊断,/v1/code-analysis/errors/parse 解析报错堆栈;调用链图谱(call-graphs nodes/edges/subgraph)与流量分析交叉定位可疑变更;对持久化的流量图可再触发 /v1/code-analysis/traffic/graphs/{id}/analyze 生成 LLM 分析报告。
Case 5 · 工单分诊与摘要(工单运营) — TicketSummary 页面对入站工单自动分类、路由与结构化摘要,把「读工单」从分钟级压到秒级;摘要与分类结果可回写客户工单系统。
Case 6 · 集成进已有平台(平台工程) — 三条集成路径:REST API(Go 平台 ~100 条路由 + Python 运行时 14 个端点,含 SSE 流式)、gRPC SelectorService(把智能路由作为共享服务嵌入自有调度链)、MCP 适配器(把 ResolveAgent 能力暴露为 Model Context Protocol 工具,供 Claude/通义等 Agent 直接调用);另有 Dify / LangGraph 集成适配层。无后端环境时可用 WebUI 演示模式(mock 数据驱动)先行体验。
Quick Start
# 1. Clone the repository
git clone https://github.com/ai-guru-global/resolve-agent.git
cd resolve-agent
# 2. Start the entire stack with one command
# (deps containers + Go platform + Python runtime + WebUI)
./scripts/start-local.sh all
# 3. Check service health
./scripts/start-local.sh status
After startup, open the WebUI at http://localhost:5174 and the Platform API at http://localhost:8080/api/v1/health.
⚠️ First run of
start-local.sh runtimeautomatically createspython/.venvand installs dependencies (.[rag]); this may take a few minutes.
Environment Requirements
| Dependency | Version | 说明 |
|---|---|---|
| Go | >= 1.25 | go.mod 锁定 1.25.0 |
| Python | >= 3.11 | 运行时(开发环境已验证 3.14) |
| Node.js | >= 20 | WebUI 开发(已验证 v22) |
| Docker | >= 20.10 | 依赖服务容器化 |
| Make | latest | 构建工具链 |
| uv | 推荐 | Python 依赖管理(无则回退 pip) |
Startup endpoints
| 服务 | 地址 | 说明 |
|---|---|---|
| WebUI | http://localhost:5174 | React 可视化控制台 |
| Platform API | http://localhost:8080 | Go 平台服务(健康检查 /api/v1/health) |
| Runtime API | http://localhost:9091 | Python Agent 运行时(/health) |
| gRPC | localhost:9090 | Platform gRPC 服务(仅健康检查与反射;业务 API 走 REST) |
| PostgreSQL | localhost:5432 | 长期记忆 / 业务数据 |
| Redis | localhost:6379 | 短期记忆 / 缓存 |
| NATS | localhost:4222 | 消息总线 |
| Milvus | localhost:19530 | RAG 向量存储 |
Step-by-Step & Makefile
./scripts/start-local.sh deps # 仅启动依赖容器 (PostgreSQL / Redis / NATS / Milvus / etcd)
./scripts/start-local.sh platform # 仅启动 Go 平台服务(自动编译)
./scripts/start-local.sh runtime # 仅启动 Python Agent 运行时(自动重建 venv)
./scripts/start-local.sh web # 仅启动 WebUI 开发服务器
./scripts/start-local.sh stop | restart | logs web | doctor
make setup-dev # 初始化开发环境(依赖 + Git Hooks)
make build # 构建 Go / Python / WebUI
make compose-up # Docker Compose 全栈启动
make test # 全量测试(Go + Python + WebUI)
make lint # 全语言 lint 门禁(Go/Python/Web/Proto)
Python Agent Example
from resolveagent.selector.selector import IntelligentSelector
from resolveagent.memory import HierarchicalMemory
from resolveagent.planning import HybridPlanner, PlanningMode
selector = IntelligentSelector(strategy="hybrid")
memory = HierarchicalMemory(session_id="session-1")
planner = HybridPlanner()
decision = await selector.route(input_text="分析代码中的安全漏洞", agent_id="my-agent")
# → RouteDecision(route_type="code_analysis", confidence=0.85)
plan = await planner.create_plan(goal="诊断 API 500 错误", mode=PlanningMode.DELIBERATIVE)
# → Plan(steps=[Step(gather_info), Step(diagnose), Step(fix)])
memory.add("user", "我想部署到 k8s", importance=0.8)
recent = memory.get_recent(limit=10)
遇到启动问题?见 Troubleshooting。
Architecture Overview
┌──────────────────────────────────────────────────────────────────────────┐
│ RESOLVEAGENT │
│ │
│ ┌──────────┐ ┌──────────────┐ ┌─────────────────────────────────┐ │
│ │ Client │───▶│ Higress │───▶│ Platform (Go) │ │
│ │ CLI/WebUI│ │ AI Gateway │ │ Registry │ Auth │ Route │ Store │ │
│ │ Mobile │ │ │ │ Feedback │ CircuitBreaker │ │
│ └──────────┘ └──────────────┘ └────────────┬──────────────────┘ │
│ │ HTTP/SSE │
│ ▼ │
│ ┌─────────────────────────────────────────┐ │
│ │ Agent Runtime (Python) │ │
│ │ │ │
│ │ ┌──────────┐ ┌──────────┐ ┌─────────┐ │ │
│ │ │ Selector │ │ Planner │ │ Memory │ │ │
│ │ │ (路由) │ │ (规划) │ │ (记忆) │ │ │
│ │ └──────────┘ └──────────┘ └─────────┘ │ │
│ │ ┌──────────┐ ┌──────────┐ ┌─────────┐ │ │
│ │ │ FTA │ │ Skills │ │ RAG │ │ │
│ │ │ 引擎 │ │ +沙箱 │ │ Pipeline│ │ │
│ │ └──────────┘ └──────────┘ └─────────┘ │ │
│ │ ┌───────────────────────────────────┐ │ │
│ │ │ Resilience │ MessageBus │ LLM │ │ │
│ │ └───────────────────────────────────┘ │ │
│ └──────────────────┬──────────────────────┘ │
│ │ │
│ ┌────────────────────────────────────────────▼──────────────────────┐ │
│ │ Data Layer: PostgreSQL │ Redis │ NATS │ Milvus/Qdrant │ │
│ └───────────────────────────────────────────────────────────────────┘ │
└──────────────────────────────────────────────────────────────────────────┘
三层分工: Go 平台负责注册表、鉴权、存储与反馈闭环等高并发基础服务;Python 运行时承载 Agent 编排与 AI 引擎(agentscope);React WebUI 面向运维操作者。三者经 HTTP/SSE 互联。
Twelve Architecture Highlights
| # | 特性 | 模块 | 说明 |
|---|---|---|---|
| 1 | Intelligent Selector | selector/selector.py |
三阶段元路由:意图分析 → 上下文增强 → 路由决策 |
| 2 | Decision Audit Logger | selector/audit.py |
完整路由决策审计追踪(异步写入,不阻塞主流程) |
| 3 | Hierarchical Memory | memory.py |
Working / Episodic / Long-term 三层记忆(TTL + LRU) |
| 4 | Hybrid Planner | planning.py |
REACTIVE + DELIBERATIVE 双模式,LLM 分解 + 规则回退 |
| 5 | ToolHub | toolhub.py |
工具发现 + Schema 注册 + Capability 映射 + 安全审计 |
| 6 | Resilience | resilience.py |
CircuitBreaker 三态熔断 + FallbackCascade 多级降级 |
| 7 | AgentMessageBus | message_bus.py |
Pub/Sub + Request/Response 消息总线 |
| 8 | FTA Engine | fta/ |
五种门类型(PRIORITY_AND 带动态时序语义)+ 最小割集 + 蒙特卡洛仿真 |
| 9 | Resilient Selector | selector/resilient_selector.py |
反馈驱动自适应路由:失败重试 + 上下文重增强 + 错误分类路由偏好 |
| 10 | Loop Engineering | pkg/feedback/ + fta/feedback_loop.py |
Observe-Orient-Decide-Act 持续反馈闭环 |
| 11 | Circuit Breaker (Go) | pkg/circuitbreaker/ |
三态熔断器自愈运维:Closed → Open → HalfOpen → Closed |
| 12 | Adaptive Weight Adjuster | selector/resilient_selector.py |
基于反馈的路由权重动态调整 + 时间衰减 + 自动降级 |
📖 逐模块代码级讲解见 Architecture Deep Dive。
CLI
resolveagent 命令行工具(Go 实现,位于 internal/cli/ + cmd/resolveagent-cli/),覆盖平台全部核心操作,并提供交互式 TUI 控制台(internal/tui/):
resolveagent agent create|list|describe|run|update|delete|logs # Agent 管理
resolveagent skill list|info|install|test|remove|init # 技能管理
resolveagent workflow create|list ... # 工作流管理
resolveagent rag ... / corpus ... # RAG 与语料
resolveagent config ... / serve / dashboard # 配置 / 本地服务 / TUI
📖 完整命令参考见 CLI 参考。
WebUI
访问 http://localhost:5174 查看可视化控制台(23 个功能页面):Dashboard / Playground(Agent 多轮对话)/ Selector 路由可视化 / FTAEngine 故障树编辑器 / Workflows / Agents / Skills / RAG / CodeAnalysis(调用链图谱)/ TicketSummary / Solutions / Monitoring / Traces / Database / Settings 等。
演示模式: 后端不可用时自动回落到浏览器端 mock 数据(按域拆分于 web/src/api/mock/),并显示演示模式标识徽标;演示数据带新鲜度测试,防止随时间推移变成"假旧"。
Configuration
平台与运行时配置统一使用 RESOLVEAGENT_* 前缀(完整清单见 .env.example 与 配置指南):
# 服务地址
RESOLVEAGENT_HTTP_ADDR=:8080
RESOLVEAGENT_GRPC_ADDR=:9090
# 数据层
DATABASE_URL=postgres://resolveagent:resolveagent@localhost:5432/resolveagent?sslmode=disable
RESOLVEAGENT_REDIS_ADDR=localhost:6379
RESOLVEAGENT_NATS_URL=nats://localhost:4222
# LLM (多 Provider:Qwen / 文心 / 智谱 / Kimi / MiMo 等 OpenAI 兼容接入)
RESOLVEAGENT_LLM_QWEN_API_KEY=your-qwen-key
KIMI_API_KEY=your-kimi-key
# 网关与可观测性
RESOLVEAGENT_GATEWAY_ENABLED=true
RESOLVEAGENT_TELEMETRY_ENABLED=true
RESOLVEAGENT_TELEMETRY_OTLP_ENDPOINT=http://localhost:4318
| 文件 | 说明 |
|---|---|
configs/resolveagent.yaml |
Go 平台服务配置(注册表/网关/存储) |
configs/runtime.yaml |
Python 运行时配置(Agent/记忆/技能) |
configs/models.yaml |
LLM 模型路由与参数配置 |
configs/examples/ |
各环境配置示例(agent / skill / FTA workflow) |
Documentation
| 类别 | 入口 |
|---|---|
| 架构深潜(代码级) | docs/architecture-deep-dive.md |
| 功能状态与已知待办 | docs/feature-status.md |
| 中文技术文档(25 篇) | docs/zh/INDEX.md — 快速开始 / 架构详解 / 智能选择器 / FTA 引擎 / RAG 管道 / 技能系统 / 数据库 Schema / 配置指南 / CLI 参考 / 部署指南… |
源码蒸馏设计文档(19 篇,带 file:line 锚点) |
docs/design/INDEX.md |
| 疑难排查 | docs/troubleshooting.md |
| 文档站点(Docusaurus) | docs-site/ —— 仅站点壳与导航;内容唯一源在 docs/ |
Testing
make test # 全量测试(Go + Python + WebUI 并行)
make test-e2e # Go E2E 测试(构建标签隔离,支持 E2E_BASE_URL 覆盖)
make lint # 全语言 lint 门禁
hack/quality-gate.sh # 质量门禁(lint + test + coverage)
- Python — 480+ 单元测试项 + 8 套集成测试(Dify 插件 / FTA 并行 / HTTP 服务 / MCP 适配器 / 弹性选择器等)
- Go —
-race+ 覆盖率;E2E 覆盖 agent lifecycle / workflow execution / feedback loop - Web — Vitest 113 项(含演示数据新鲜度守卫)
- 契约 — 以
router.go的 95 条路由注册为唯一事实源的 OpenAPI 契约测试 - 工程治理 — golangci-lint 全量规则清零或显式注明理由;sentinel 错误链 + 防泄漏统一错误出口
Observability
- Prometheus 指标 —
resolveagent_selector_decisions_total/resolveagent_audit_records_total/resolveagent_circuit_breaker_state/resolveagent_feedback_signals_total/resolveagent_adaptive_selector_weights等(pkg/telemetry/) - OpenTelemetry — 结构化日志带 trace 关联(
pkg/logger/),OTLP 导出(RESOLVEAGENT_TELEMETRY_OTLP_ENDPOINT) - 健康检查 — liveness / readiness 端点(
pkg/health/):Platform/api/v1/health,Runtime/health - 审计追踪 — Selector 决策审计(异步落库)+ ToolHub 执行审计 + 熔断器状态变化经反馈 Observer 采集
Feature Status
核心引擎与基础设施的就绪状态、以及经代码实测裁定的已知待办,统一维护在 docs/feature-status.md。
Project Structure
resolve-agent/
├── api/ # 协议定义 (OpenAPI / JSON Schema / proto)
├── cmd/
│ ├── resolveagent-cli/ # CLI 应用入口
│ └── resolveagent-server/ # Platform Server 入口
├── internal/
│ ├── cli/ # CLI 实现 (agent/skill/workflow/rag/corpus/config)
│ └── tui/ # 交互式 TUI 控制台
├── pkg/ # Go 平台服务
│ ├── circuitbreaker/ # 🔄 三态熔断器
│ ├── config/ errors/ event/ # 配置 / sentinel 错误链 / 事件体系
│ ├── feedback/ # 🔄 反馈循环 (Collector/RingBuffer/Aggregator)
│ ├── gateway/ health/ logger/
│ ├── registry/ # 14 个领域 Registry (agent/skill/rag/fta/solution/traffic…)
│ ├── retry/ server/ service/ store/ telemetry/ version/
├── python/src/resolveagent/
│ ├── selector/ # 🧠 智能选择器 (意图/上下文/路由/审计/弹性)
│ ├── memory.py planning.py toolhub.py resilience.py message_bus.py
│ ├── fta/ # 🔍 FTA 引擎 + 反馈循环 + 回归验证
│ ├── rag/ # RAG 管道 (Milvus/Qdrant 双后端)
│ ├── skills/ llm/ mcp/ hooks/ code_analysis/ docsync/ agent/ runtime/
├── web/ # 🌐 React WebUI (Vite + Tailwind + shadcn/ui)
├── benchmarks/wiki-k8s/ # 基准语料(K8s wiki)
├── docs/ # 文档 (架构深潜/功能状态/设计蒸馏/中文文档)
├── docs-site/ # 📚 Docusaurus 站点壳(内容引用 docs/)
├── configs/ # 运行配置 (resolveagent/runtime/models + examples)
├── scripts/ # 开发运维脚本 (start-local.sh 一键启动)
├── deploy/ # 部署配置 (docker / docker-compose / helm)
├── skills/ # 技能注册 + 示例技能 (k8s-pod-crash / ticket-handler…)
├── examples/ # 示例 (quickstart / integrations / mobile-demo)
├── integrations/dify/ # Dify 集成
├── test/ # E2E (Go 构建标签) / 集成 / fixtures
├── hack/ # 开发工具 (quality-gate / coverage-report)
└── .github/workflows/ # CI/CD (ci / e2e / release / codeql)
Deployment
make docker # 构建镜像 (platform / runtime / webui)
make compose-up # Docker Compose 全栈
make helm-install # Helm 部署
make migrate-up # 数据库迁移
make seed # 种子数据
生产部署推荐使用 deploy/helm/resolveagent/(含健康检查、资源限制),详见 部署指南。
Roadmap
完整路线图(唯一权威副本)见 ROADMAP.md,实时状态见 GitHub Issues。
| Version | Theme | Highlights |
|---|---|---|
| v0.1.0 ✅ | Foundation | Go platform (gRPC + REST), Python runtime, FTA engine, Intelligent Selector, RAG, WebUI, CLI, Docker Compose, Helm. |
| v0.2.0 ✅ | Hardening | DB migrations, unified error handling, OpenTelemetry logging, health checks, integration tests, retry with backoff. |
| v0.3.0 ✅ | Quality & Foundation | CI/CD, unified versioning, router refactoring (2160 行 → 15 个域文件), PostgreSQL Registry 持久化, MCP adapter. |
| v0.4.0 ✅ | Core Capability | FTA 并行求值, 多语言代码分析 (tree-sitter), LangGraph 集成, Dify 插件, 语料导入链路. |
| v0.5.0 🔜 | Enterprise Readiness | Multi-tenant, RBAC, comprehensive audit logging, SSO (OIDC/SAML), data encryption, compliance. |
| v0.6.0 | Ecosystem & Scale | Skill marketplace, plugin SDK, horizontal agent runtime, distributed workflows, NATS JetStream, advanced RAG. |
Long-term vision: multi-cloud and edge deployment, visual workflow designer, AI-powered observability and self-healing.
Contributing
We welcome contributions! Please read CONTRIBUTING.md for:
- Development environment setup (
make setup-dev) - Branch naming conventions (
feat/,fix/,docs/, etc.) - Conventional Commits format
- Pre-commit hooks and linting (
make lint,make fmt) - Pull request process and code review requirements
- Coding standards for Go, Python, and TypeScript/React
By contributing, you agree that your contributions will be licensed under the Apache License 2.0.
Security
If you discover a security vulnerability, please do not open a public issue. Instead, report it privately via GitHub Security Advisories or contact the maintainers listed in MAINTAINERS.md.
Built-in security practices:
- ToolHub SecurityPolicy — 技能/工具基于角色的权限控制 + 执行审计
- Skill Sandbox — 技能在沙箱中执行,隔离宿主环境
- 防泄漏错误出口 — registry 错误统一经
writeRegistryError输出,未知错误码不回显内部消息 - Higress 网关鉴权 — 可选 JWT 鉴权(
RESOLVEAGENT_GATEWAY_AUTH_ENABLED)
Community & Support
- 💬 GitHub Discussions — Ask questions, share ideas, and connect with other users.
- 🐛 GitHub Issues — Report bugs or request features.
- 📧 Maintainers — See MAINTAINERS.md for contact information.
- ⭐ Star us on GitHub — If ResolveAgent is helpful, please consider starring the repository to help others discover it.
Citation
If you use ResolveAgent in your research or production work, please consider citing it:
@software{resolveagent,
title = {ResolveAgent: A Production-Grade AIOps Mega-Agent Platform},
author = {The ResolveAgent Contributors},
url = {https://github.com/ai-guru-global/resolve-agent},
year = {2026},
license = {Apache-2.0}
}
Acknowledgments
ResolveAgent builds on the shoulders of excellent open-source projects and communities:
- AgentScope — Agent 编排
- Higress — AI 网关
- LangGraph — 图编排参考
- MCP — Model Context Protocol
- Milvus / Qdrant — 向量数据库
- Vite / React / Tailwind CSS — WebUI 技术栈
- Go / Python / Node.js — 核心运行时
A special thanks to all contributors who have helped shape ResolveAgent.
License
Apache 2.0 License — see LICENSE