# ApeMind — Full Product, Comparison & Security Reference Canonical site: https://apemind.ai/ Index file: https://apemind.ai/llms.txt Sitemap: https://apemind.ai/sitemap.xml Open-source core engine (ApeRAG): https://github.com/apecloud/ApeRAG ## 1. Product Positioning ApeMind is an enterprise RAG, knowledge base, and AI agent platform. It ingests enterprise documents, websites, databases, and multimodal files into multi-index knowledge bases (vector, full-text, and optional governed knowledge graph), enables agents to retrieve evidence, inspect original passages, and invoke authorized external tools, and closes the quality loop with execution diagnostics, versioned Q&A datasets, and immutable shareable evaluation reports. Delivery options: - SaaS: https://apemind.ai/ - Private deployment (single VM): Docker Compose (single-host fault domain; software fee covers ApeMind only) - Private deployment (HA cluster): Kubernetes + KubeBlocks Enterprise (3+ servers, multi-replica Frontend/API/Index Workers, managed PostgreSQL, Graph PG, Redis, Elasticsearch, Qdrant, and standalone RustFS object storage by default or legacy KubeBlocks MinIO when explicitly enabled) ## 2. Source-Verified Capability Matrix | Dimension | Verified ApeMind Capability | Documented Scope & Boundary | Reference | | --- | --- | --- | --- | | Ingestion & Parsing | PDF, Office docs, full-site web crawl, URLs, plain text, images, and audio/video; configurable parsers including MinerU, Mistral OCR, LlamaParse, and MarkItDown. | OCR accuracy and external data flow depend on which parser service is configured in your deployment. | https://apemind.ai/docs/concepts/pdf-parsing.html | | Chunking Control | Structure-aware block packing, type-specific splitting, Jev decision-model topic boundary suggestions, and interactive visual chunk boundary editing. | AI boundary suggestions are reviewed individually before saving; re-indexing applies after confirmation. | https://apemind.ai/docs/guide/chunk-boundaries.html | | Hybrid & Graph Retrieval | Parallel Vector (Qdrant) + Full-Text (Elasticsearch) + optional Knowledge Graph (Graph PostgreSQL) retrieval fused by Reciprocal Rank Fusion (RRF), plus governed entity merge/noise rules with revert history. | Knowledge Graph v2 topic discovery (Louvain communities) is a development preview; graph indexing adds model cost and does not guarantee better accuracy on every query. | https://apemind.ai/docs/concepts/indexes.html | | Citations & Evidence | Answers cite only passages the agent actually read during that turn, tracing back to chunk, source document, and collection. | A citation is a path back to source evidence for human verification, not automatic proof that an LLM conclusion is error-free. | https://apemind.ai/docs/concepts/citations.html | | External Connections | Organization-managed Connections for PostgreSQL, MySQL, Feishu / Lark, IMAP email, GitHub, remote Streamable HTTP MCP servers, and HTTP/OpenAPI 3.0–3.1 APIs, plus automated collection sync. | DB connections enforce read-only AST + read-only transaction (max 200 rows, 256 KiB, 30s) and still require a dedicated read-only DB user; credentials are stored in plaintext in DB (see Security page). | https://apemind.ai/docs/guide/connections.html | | Protocols & Interfaces | REST API with OpenAPI/Swagger, built-in MCP Server (/mcp/), OpenAI Chat Completions compatible endpoint (/v1/chat/completions), A2A 1.0 JSON-RPC/SSE (/v1/a2a/bots/{bot_id}), declarative config API, Desktop app, and standalone CLI. | A2A v1 supports single-turn async tasks (no multi-turn follow-up or push notifications inside one task) and uses private API-key-protected Agent Cards. | https://apemind.ai/docs/developers/overview.html | | Evaluation & Diagnostics | Per-turn execution diagnostics (tool calls, arguments, errors, context usage), versioned Q&A datasets, baseline-vs-candidate evaluation comparisons, and immutable shareable report snapshots. | Unattended self-improvement loops require external orchestration (e.g., CLI + your CI/workflow); not a built-in one-click autonomous RSI switch. | https://apemind.ai/docs/guide/datasets.html | | Multi-Tenancy & RBAC | Separate Site roles (admin, rw, ro) and Organization roles (Owner, Admin, Editor, Viewer + custom action roles); workspace-bound normal, managed agent, and service-account API keys; org quotas & rate limits. | Site admins manage platform governance but do not automatically gain read access to organization knowledge bases or agents. | https://apemind.ai/docs/concepts/permissions.html | | Private Deployment | SaaS (apemind.ai), single-VM Docker Compose, and multi-replica Kubernetes + KubeBlocks Enterprise (3+ servers) managing PostgreSQL, Graph PG, Redis, Elasticsearch, Qdrant, and RustFS object storage. | Docker Compose has a single-host fault domain; private deployment does not automatically make external model, parser, or search providers offline. | https://apemind.ai/docs/admin/deployment.html | ## 3. Comparison Positioning (RAG & Knowledge Base Ecosystem) Note: External platforms iterate frequently. Treat the summaries below as architectural positioning and run a side-by-side POC with the same documents and identical model versions before choosing a platform. Full bilingual comparison pages: - English: https://apemind.ai/compare/en - Chinese: https://apemind.ai/compare/zh ### ApeMind (Enterprise Knowledge Base & Agentic RAG Platform) - Best fit: Teams that need verifiable cross-document investigation, hybrid vector + full-text + governed knowledge graph retrieval, org-level access boundaries, versioned evaluation datasets, and private deployment. - Architectural focus: Focuses on the full evidence chain—from layout parsing and editable chunk boundaries to read-verified citations, execution diagnostics, and KubeBlocks/Compose private operations—rather than a general no-code consumer bot canvas. ### Dify (Visual LLM Workflow & App Builder) - Best fit: Teams building diverse LLM applications quickly via drag-and-drop workflow graphs, prompt templates, and a broad marketplace of model and tool plugins. - Architectural focus: Dify centers on visual workflow orchestration across many app types; ApeMind centers on deep knowledge-base retrieval (including governed GraphRAG), org-scoped Connections, and evaluation-driven agent iteration. ### RAGFlow (Open-Source Document Parsing & RAG Engine) - Best fit: Teams looking for an open-source self-hosted RAG engine with deep document layout parsing and chunking templates. - Architectural focus: Both invest heavily in document parsing and verifiable retrieval. ApeMind pairs pluggable parsers (MinerU, Mistral OCR, LlamaParse, MarkItDown) with editable chunk boundaries, governed knowledge graphs, MCP/A2A protocols, and versioned agent evaluation reports. ### FastGPT (Open-Source Knowledge Q&A & Workflow Platform) - Best fit: Teams that want approachable open-source setup for standard knowledge-base Q&A and visual flow orchestration. - Architectural focus: FastGPT emphasizes fast onboarding for Q&A and flow-based bots; ApeMind emphasizes multi-lane RRF retrieval (vector + full-text + graph), dual-plane site/org RBAC, workspace-bound service keys, and KubeBlocks Enterprise HA operations. ### HiAgent & Coze (Cloud / Enterprise Low-Code Agent Platforms) - Best fit: Organizations prioritizing low-code agent creation, built-in channel distribution, and managed ecosystem plugins. - Architectural focus: Coze and HiAgent optimize for ecosystem-integrated bot publishing and low-code app assembly. ApeMind targets teams that require direct control over storage topology, model/parser routing, open protocols (MCP, OpenAI-compat, A2A 1.0), and self-hosted data boundaries. ### Tencent IMA (腾讯 IMA) (Personal & Lightweight Team Knowledge Assistant) - Best fit: Individuals and small groups collecting articles, notes, and files inside the WeChat/Tencent ecosystem for everyday Q&A. - Architectural focus: Tencent IMA is designed as a lightweight personal/team knowledge companion; ApeMind is built for private enterprise deployments with organization RBAC, external database/Feishu/MCP connections, OpenAPI/SDK integration, and auditable agent evaluations. ### WeKnora (Open-Source Modular RAG Framework) - Best fit: Engineering teams evaluating lightweight, modular open-source RAG pipelines for document indexing and Q&A. - Architectural focus: WeKnora focuses on a modular RAG pipeline; ApeMind ships as a full enterprise platform adding governed entity graphs, managed external Connections, CLI/declarative config workflows, execution traces, and formal Kubernetes HA deployment. ## 4. Verifiable Security & Architecture Facts Full bilingual security & architecture pages: - English: https://apemind.ai/security/en - Chinese: https://apemind.ai/security/zh | Area | Verified Behavior | Explicit Boundary / Operator Responsibility | Reference | | --- | --- | --- | --- | | Deployment Topologies | Offered as SaaS (apemind.ai) or two official private deployment architectures: single-VM Docker Compose and Kubernetes + KubeBlocks Enterprise (3+ servers with multi-replica Frontend, API, and Index Workers). | Docker Compose operates within a single-host fault domain. Kubernetes HA requires KubeBlocks Enterprise for managed database lifecycle, backups, and failover. | https://apemind.ai/docs/admin/deployment.html | | Data Storage Locations | Raw files are stored in object storage (standalone RustFS chart by default on new private installs, or legacy KubeBlocks MinIO when explicitly enabled); metadata in PostgreSQL; knowledge graph in Graph PostgreSQL; vectors in Qdrant; full-text index in Elasticsearch; queues and cache in Redis. | Operators own infrastructure disks, backup retention, restoration drills, and cluster access policies in private deployments. | https://apemind.ai/docs/admin/deployment.html | | Network & Outbound Boundaries | In private deployment, core application and data services stay inside your VPC/cluster. Model endpoints, parsers (MinerU, Mistral OCR, LlamaParse), web search/read, and external Connections are explicitly configured per deployment or organization. | Private deployment does not automatically make all integrations offline. Audit every configured LLM, embedding, rerank, OCR, web-access, and MCP/HTTP endpoint before handling sensitive documents. | https://apemind.ai/docs/admin/models.html | | Dual-Plane Role Isolation | Site roles (admin, rw, ro) govern platform operations, users, quotas, and rate limits. Organization roles (Owner, Admin, Editor, Viewer, plus custom action-level roles) govern knowledge bases, agents, datasets, and connections. | Site admins do not automatically gain access to organization business data; reading an organization collection still requires membership and matching action permission. | https://apemind.ai/docs/concepts/permissions.html | | Default Deny & Existence Masking | All entry points (Web UI, REST API, Agents, MCP, Share links) share one authorization kernel evaluated on every real read. Unlisted actions are denied by default. | Org members lacking an action receive explicit denial; callers unrelated to a resource receive "not found" so unauthorized users cannot probe resource existence. | https://apemind.ai/docs/concepts/permissions.html | | API Keys & Machine Identities | Every API key is pinned to a single workspace at creation and never floats when the holder switches workspaces. Supports Normal keys (30/90/365-day expiry), Managed agent keys (lifecycle-bound, protected from manual deletion), and Service Account keys (non-human org credentials, plaintext shown once). | Data API keys cannot invoke platform admin endpoints. Org governance lists all keys pinned to the org and allows revoking unprotected keys. | https://apemind.ai/docs/concepts/permissions.html | | External Connection Credentials | Connection credentials and upstream OAuth tokens are write-only (never returned in API responses, logs, audit entries, or errors) and stored in a dedicated table separate from non-secret config. Changing target host, URL, or auth mode clears stored credentials automatically. | Documented limitation: connection credentials and upstream OAuth tokens are stored in plaintext in PostgreSQL (not encrypted at rest). Database backups include them and must be protected and encrypted at the storage/backup layer. | https://apemind.ai/docs/guide/connections.html | | Database Connection Safeguards | PostgreSQL and MySQL connections apply read-only SQL AST filtering (allowing SELECT / read-only CTE / UNION; blocking DML, DDL, multi-statement, COPY, EXPLAIN, SET, system catalogs, pg_sleep, dblink, LOAD_FILE) and execute inside read-only transactions capped at 200 rows, 256 KiB, and 30s. | Application guards do not replace database-side least privilege: always create a dedicated read-only database user. Grant connection.create only to trusted roles because test-connection probes internal DB ports. | https://apemind.ai/docs/guide/connections.html | | Protocol Gates (MCP & A2A) | MCP Server (/mcp/) enforces layered workspace_access, collection_access, and web_access gates and never falls back to browser cookies or server env keys. A2A 1.0 Agent Cards (/v1/a2a/bots/{bot_id}/agent-card.json) are private and require a workspace-bound API key. | ApeMind does not expose a public /.well-known/agent-card.json endpoint; unauthorized A2A callers receive HTTP 404. | https://apemind.ai/docs/developers/a2a.html | | Quotas, Rate Limits & Audit | Supports organization quotas (collections, total docs, docs per collection, bots), system and per-user rate-limit policies with optimistic revision locking (409 REVISION_CONFLICT), 1–168h rejection records, and API/tool audit logs. | Quotas are enforced per organization, not per individual user. Rejection logs explain HTTP 429 events and complement—rather than replace—the API/MCP audit ledger. | https://apemind.ai/docs/admin/governance.html | | Open-Source Auditability | The open-source core codebase (ApeRAG) is publicly available on GitHub for code inspection; ApeMind Enterprise builds commercial governance, enterprise connectors, and managed private delivery on top. | Verify whether a specific enterprise feature lives in open-source ApeRAG or the commercial ApeMind Enterprise distribution during procurement. | https://apemind.ai/docs/start/what-is-apemind.html | ## 5. CLI & Desktop Installation Installing the standalone CLI does not deploy a local ApeMind server. Full CLI guide: https://apemind.ai/download/cli.md Download page: https://apemind.ai/download Release metadata JSON: https://apemind.ai/api/v2/public/desktop-release - macOS / Linux install: `curl -fsSL https://apemind.ai/ap/install.sh | APEMIND_VERSION=latest sh` - Windows PowerShell install: `$env:APEMIND_VERSION = "latest"; Invoke-RestMethod https://apemind.ai/ap/install.ps1 | Invoke-Expression` - Verify install: `apemind --help` - Login: `apemind auth login` - Discover CLI commands: `apemind capabilities --json` ## 6. Bilingual Topic Pages - Enterprise RAG & Private Deployment: https://apemind.ai/enterprise/en | https://apemind.ai/enterprise/zh - Knowledge Graph & Topic Discovery: https://apemind.ai/knowledge-graph/en | https://apemind.ai/knowledge-graph/zh - Agent RSI (Execution Diagnostics & Evaluation): https://apemind.ai/rsi/en | https://apemind.ai/rsi/zh - RAG & Knowledge Base Comparison: https://apemind.ai/compare/en | https://apemind.ai/compare/zh - Security & Deployment Architecture: https://apemind.ai/security/en | https://apemind.ai/security/zh ## 7. Complete Documentation Directory (Chinese) Documentation home: https://apemind.ai/docs/index.html ### 使用 (https://apemind.ai/docs/guide/index.html) - [认识 ApeMind](https://apemind.ai/docs/start/what-is-apemind.html): 它解决什么问题、核心能力如何组织,以及四大核心概念。 - [十分钟跑通第一个问答](https://apemind.ai/docs/start/quickstart.html): 建库、上传、配置 Agent、提问与句级溯源核对,一次走完。 - [添加文档与批量导入](https://apemind.ai/docs/guide/upload.html): 本地文件、文件夹、网页 URL、整站爬取与粘贴文本,一站导入知识库。 - [数据源自动同步](https://apemind.ai/docs/guide/data-sources.html): 对接飞书、网站与邮箱实现定时增量同步,文档更新不再人工搬运。 - [数据库知识库与自然语言问数](https://apemind.ai/docs/guide/database-kb.html): 连接只读业务库、自动 Schema 发现、保留人工业务注释与结构变更审核。 - [解析预览与调整文档切分](https://apemind.ai/docs/guide/chunk-boundaries.html): 查看解析与分块效果,手动微调切分线或用决策模型自动切分。 - [知识图谱探索与治理](https://apemind.ai/docs/guide/knowledge-map.html): 通过主题气泡发现跨文档关联,合并重复实体、屏蔽噪声并定制领域提示词。 - [导出与备份知识库](https://apemind.ai/docs/guide/export.html): 将知识库原始文件与清单打包为 ZIP 归档、合规备份或跨环境迁移。 - [知识市场](https://apemind.ai/docs/guide/marketplace.html): 发布、发现与只读订阅高质量公共知识库,一次维护多团队复用。 - [配置 Agent:模型、代码模式与权限](https://apemind.ai/docs/guide/agent-config.html): 人设指令、组合模型容灾、知识库与 MCP 授权、沙箱代码模式与写操作确认。 - [对话与句级引用核对](https://apemind.ai/docs/guide/chat.html): 与 Agent 对话、@ 指定资料、句级原文溯源抽屉、反馈导出与公开分享。 - [接入 IM 机器人渠道](https://apemind.ai/docs/guide/im-channels.html): 将 Agent 接入飞书、钉钉、企业微信、Slack 与 Telegram,实现群聊 @ 问答。 - [分享、公开访问与网站嵌入](https://apemind.ai/docs/guide/agent-access.html): 开放给组织同事、生成免登录链接、嵌入网站气泡或签发专属 API 密钥。 - [组织助手](https://apemind.ai/docs/guide/org-assistant.html): Owner 与管理员在对话里盘点、诊断并经批准后自动搭建组织资源。 - [连接管理:数据库、飞书、邮箱、MCP 与 HTTP](https://apemind.ai/docs/guide/connections.html): 集中管理外部系统凭据、按操作逐条审批授权与写操作确认策略。 - [技能、函数与定时任务](https://apemind.ai/docs/guide/automation.html): 用技能沉淀专家方法论、用函数封装受管代码工具、用定时任务自动巡检。 - [问答集、标准答案与 Agent 评测](https://apemind.ai/docs/guide/datasets.html): 把高频问题绑定为知识库标准答案,并通过版本化评测对比改动、分享报告。 - [执行诊断与上下文用量](https://apemind.ai/docs/guide/agent-diagnostics.html): 追踪工具调用、排查上下文 Token 暴涨,并按轮次 ID 快速提交管理员取证。 ### 场景案例 (https://apemind.ai/docs/cases/index.html) - [企业制度问答助手](https://apemind.ai/docs/cases/policy-qa.html): 多库权限隔离、高频标准答案绑定、40 题版本化验收与飞书/内网双端上线。 - [财报与研报多表深度分析](https://apemind.ai/docs/cases/financial-report-analysis.html): PDF 复杂表格解析 + 代码模式 Python 精确计算 + 句级原文溯源审计。 - [研发与运维知识库 IM 机器人](https://apemind.ai/docs/cases/dev-kb-im-bot.html): 飞书/文档站自动同步 + 知识图谱串联服务与责任人 + 群聊值班答疑与定时巡检。 - [企业业务问数助手](https://apemind.ai/docs/cases/nl2sql-analytics.html): 数据库知识库 Schema 注释治理 + 只读权限隔离 + 代码模式可视化经营分析。 ### 原理 (https://apemind.ai/docs/concepts/index.html) - [版本化入库、增量复用与过载保护](https://apemind.ai/docs/concepts/ingestion.html): 不可变解析快照、文档微调增量复用向量与图谱、多解析器兜底,以及自适应并发限流。 - [版面理解与高精度 PDF 解析](https://apemind.ai/docs/concepts/pdf-parsing.html): 原生文字层优先、按栏还原阅读顺序、零 OCR 幻觉表格重建与扫描页按需识别。 - [结构感知与 Jev 决策模型切分](https://apemind.ai/docs/concepts/chunking.html): 结构块优先、跨片段层级上下文前缀,以及专用小模型驱动的全局动态规划切分。 - [三路混合检索与跨模态融合](https://apemind.ai/docs/concepts/indexes.html): 向量、全文与图谱协同,加上实体别名扩展、桥接块挽救、热点降噪与 RRF 融合。 - [可治理的企业级知识图谱](https://apemind.ai/docs/concepts/graph.html): 从原文血缘、安全删改到实体去重合并与可控构建成本,让 GraphRAG 真正进入生产。 - [证据优先的 Agent 运行时](https://apemind.ai/docs/concepts/agent-loop.html): 多跳检索与先读原文、原生 MCP 工具与代码模式、自纠包装层,以及异步流式持久执行。 - [缓存友好的长对话上下文管理](https://apemind.ai/docs/concepts/context-window.html): KV Cache 前缀保护、确定性历史账本、可逆原文指针与原子轮次淘汰机制。 - [可核验的句级引用与原文溯源](https://apemind.ai/docs/concepts/citations.html): 先读原文再引用、版本锁定页码与图表、回读实时权限校验与分享脱敏冻结快照。 - [多租户权限隔离与零绕过安全边界](https://apemind.ai/docs/concepts/permissions.html): 零绕过站点边界、防存在性枚举、Prompt 注入越权隔离与人员/服务账号交接设计。 ### 开发者 (https://apemind.ai/docs/developers/index.html) - [接入选型与快速开始](https://apemind.ai/docs/developers/overview.html): 五大接入入口对比、基础地址约定与 5 分钟首个调用闭环。 - [认证、API Key 与工作区上下文](https://apemind.ai/docs/developers/auth.html): API Key、OAuth 与服务账号选型、X-ApeMind-Workspace-Id 与密钥轮换。 - [声明式配置接口](https://apemind.ai/docs/developers/declarative-config.html): 用 metadata + spec 声明式管理 Agent、模板与知识库,支持局部更新、预演与状态诊断。 - [对话与异步 Turn 集成](https://apemind.ai/docs/developers/chat-api.html): 异步 Turn 状态机、幂等键、SSE 订阅、交互确认钩子以及附件/导出/分享规则。 - [知识库与文档生命周期](https://apemind.ai/docs/developers/collections-api.html): 两步上传确认与批量导入、多通道索引状态机、失败重建、异步导出与市场边界。 - [统一连接与外部工具接入](https://apemind.ai/docs/developers/connections-api.html): 通过统一 Connections API 注册并授权外部 MCP、数据库与 HTTP 连接。 - [问答集评测与 CLI 调优闭环](https://apemind.ai/docs/developers/datasets-api.html): ETag 并发契约、两步批量导入、外部评测回传与 apemind CLI 自动迭代闭环。 - [Agent 访问点、公开链接与网站嵌入](https://apemind.ai/docs/developers/agent-access-api.html): Agent 专属 API Key 三层作用域、免登录公开访问、网站嵌入气泡与运行身份边界。 - [MCP Server 接入与工具参考](https://apemind.ai/docs/developers/mcp-server.html): 在 Cursor、Claude Code 或自定义客户端接入 ApeMind MCP,含完整工具清单与调用契约。 - [OpenAI 兼容接口](https://apemind.ai/docs/developers/openai-compat.html): 用现有 Python / TypeScript OpenAI SDK 直接调用 ApeMind Agent。 - [A2A 1.0 协议接入](https://apemind.ai/docs/developers/a2a.html): 使用私有 Agent Card、JSON-RPC 与 SSE 将 ApeMind Agent 接入多智能体编排框架。 ### API 参考 (https://apemind.ai/docs/api/index.html) ### 部署与管理 (https://apemind.ai/docs/admin/index.html) - [部署架构与实操指南](https://apemind.ai/docs/admin/deployment.html): Docker Compose 单机快速部署与 Kubernetes (Helm) + KubeBlocks Enterprise 生产部署实操。 - [容量规划、高可用与会话亲和](https://apemind.ai/docs/admin/sizing.html): 硬件规格估算、多副本统一网关会话亲和,以及反向代理 FORWARDED_ALLOW_IPS 信任规则。 - [模型管理、容灾与 Jev 决策模型](https://apemind.ai/docs/admin/models.html): 接入供应商与默认用途、搭建组合模型自动故障切换、校验 Jev 决策模型与平滑下线。 - [平台治理、沙箱与全局配置](https://apemind.ai/docs/admin/governance.html): 用户与组织配额、SSO/OAuth、企业许可证、限流、沙箱阀门、品牌定制与全局能力开关。 - [运维、告警、备份与审计](https://apemind.ai/docs/admin/operations.html): 每日数据库自动备份与恢复、通知渠道与告警规则、可观测性与索引死信、审计日志与保留期。 - [诊断与支持](https://apemind.ai/docs/admin/diagnostics.html): 只读诊断助手、按 turn_id 取证、模型连通性测试与 diagnose.sh 脱敏日志采集。 - [版本更新](https://apemind.ai/docs/admin/release-notes.html): 按版本号与关键词检索各版本新增、改进、修复与不兼容变更。