汇理极知 RatioMax · 企业级 AI 知识库与智能体平台｜支持私有化部署

RatioMax 企业级 AI 知识库与智能体平台：让文档活起来

RatioMax 将散落在网盘、文档与 IM 中的非结构化资料，自动转化为可检索、可推理、可复用的知识资产。RAG、Agent 推理与自动 Wiki 一体化——每一个回答都有据可依，每一份数据都自主可控。

  • 20+ 模型厂商兼容
  • 29 MCP 标准工具
  • 7 向量库后端扇出
  • 10+ 文档格式解析

已被这些企业选用

  • 广东中塑新材料股份有限公司 · 股票代码 301686
  • 重庆博悦机械制造有限公司 · 矿山破碎机械耐磨件

产品定位

从静态文档管理，到知识活化

传统文档系统只解决“存”与“找”。RatioMax 定位于企业知识管理的智能化升级层，让知识从静态文档，变为可被 AI 调用、可被业务流程集成的动态智能资产。

  • 知识存储：传统文档系统 — 文档上传、分类、版本管理; RatioMax · 知识活化 — 多源自动同步、多格式解析、结构化入库
  • 知识检索：传统文档系统 — 关键词全文搜索; RatioMax · 知识活化 — 语义向量 + 关键词混合检索 + 智能重排
  • 知识结构化：传统文档系统 — 无; RatioMax · 知识活化 — Agent 自治生成 Wiki 页面与知识图谱
  • 知识应用：传统文档系统 — 人工查阅; RatioMax · 知识活化 — 智能体多步推理、工具调用、多渠道服务
  • 知识服务：传统文档系统 — 无标准接口; RatioMax · 知识活化 — MCP 标准协议 · 29 工具 · 三种传输方式

核心价值

三个维度，重塑企业知识价值

知识资产化

  • 飞书知识库 / 云盘、Notion、语雀、RSS 自动定时同步，Chrome 插件网页采集，API / CLI 上传
  • PDF、Word、图片、Excel 等十余种文档格式智能解析（含 EPUB、MHTML）
  • 智能体自治生成 Wiki 页面与知识图谱，建立跨文档引用网络
  • 分块可视化编辑、版本快照、diff 对比、一键回滚与自动重建索引

服务智能化

  • ReAct（推理 + 行动）循环驱动多步任务，支持并行工具调用与思考模式
  • 语义向量 + 关键词混合检索，结合 ReRank 重排与跨库检索提升命中精度
  • 面向数据查询与分析场景的专用数据分析 Agent
  • 集成 Langfuse 追踪，覆盖从提问入口到知识检索的完整链路

集成标准化

  • MCP 标准接口：29 个工具，stdio / SSE / HTTP 三种传输，OAuth2 授权与人机审批
  • 企业微信、飞书、钉钉、云之家、Slack、Telegram 等主流 IM 渠道
  • DeepSeek、Qwen、智谱、混元、OpenAI、Gemini 等二十余家模型厂商
  • API 文件直链：第三方应用无需二次鉴权即可加载文件

数据流架构

从一份文档，到一次有据可依的回答

数据在平台内经历四个阶段，每个阶段均可观测、可排查。

  1. 文档入库：通过 API、IM、数据源连接器上传文档，系统创建入库任务并投递到消息队列。
  2. 知识结构化：解析引擎提取文本，按策略分块，向量化后写入向量库与对象存储。
  3. 知识检索：提问向量化后进行语义相似检索，结合关键词混合检索与智能重排，返回最相关内容。
  4. 知识应用：检索结果注入大模型上下文，智能体按推理—行动循环调用工具，生成带引用的回答。

核心能力

一个平台，知识库与智能体一体化

从解析、分块、检索到推理与结构化，RatioMax 在每一层都为企业场景而设计。

RAG 知识引擎：先检索企业知识，再组织回答

分块是 RAG 质量的决定性因素。RatioMax 提供多级分块策略与全生命周期管理，并以混合检索与重排保证命中精度，回答自动标注引用来源，点击即可查看原文段落。

  • 多解析引擎：OpenDataLoader 处理复杂版面 PDF，PaddleOCR-VL 处理图片与扫描件
  • 实时调试面板，调整分块参数即时预览
  • 版本快照、diff 对比与一键回滚，自定义元数据用于检索过滤
  • 跨 7 种向量库扇出：pgvector、Milvus、Qdrant、Weaviate、OpenSearch、Apache Doris、腾讯云 VectorDB

智能体与工具：会推理、会行动的企业智能体

智能体以 ReAct 循环驱动多步任务，自主规划步骤、调用工具、观察结果、调整策略；同一轮中可并行调用多个独立工具，减少推理轮次。

  • 并行工具调用，复杂任务更少轮次完成
  • 技能系统支持加载预置技能
  • 沙盒执行、范围化运行，防止越权
  • 面向数据查询与分析的专用数据分析 Agent

Wiki 知识结构化：把散乱文档，变成相互链接的知识网络

Wiki 模式是 RatioMax 的差异化能力。Agent 从原始文档自治生成结构化、相互链接的 Markdown Wiki 页面及知识图谱，而不只是简单摘要。

  • 知识提取 → 页面生成 → 跨文档引用 → 知识图谱
  • 页面快照、行级 diff、一键回滚与浏览器内手动编辑
  • 任务队列 + 死信队列，支撑万级文档批量处理
  • 实体与关系可视化，知识网络一目了然

全链路可观测：每一个回答，都可定位、可回放、可优化

集成 Langfuse 可观测平台与 W3C 跨服务追踪，从提问入口到知识检索逐环节记录；任务调度为高并发与慢任务做了专门治理。

  • 分阶段 Worker 池：解析、向量化、重排互不阻塞
  • 按模型并发治理，防止打满模型 API 速率限制
  • 失败任务进入死信队列，可查看原因、重试或丢弃
  • Prompt 缓存命中率与 Token 节省量一目了然

客户案例

已在新材料、装备制造等行业落地

成效数据为项目预期目标

广东中塑新材料股份有限公司

股票代码 301686 · 国家级专精特新“小巨人”企业 · 改性工程塑料

成立于 2009 年，专注于改性工程塑料研发、生产和销售，建有东莞长安、江西吉安、东莞水乡三大生产基地。

行业挑战：多品种、小批量、高频换料，手工编制周排产计划需 4–8 小时，异常重排需 1–3 天，高度依赖计划员个人经验。

知识管理痛点：工艺配方、设备手册、品质标准、研发记录分散在纸质文件与 Excel 中，换产查询耗时易错、品质追溯效率低。

方案应用：在洪梅工厂 APS 计划排程 AI 智能化建设项目中，RatioMax 承担企业知识库与 APS 智能问数建设，以 RAG 为核心、与 NL2SQL 双引擎融合。

应用价值 (项目预期目标)

  • ≤30 分钟 计划编制时间（原 4–8 小时/次）
  • 80%+ 计划数据查询效率提升
  • 30–45% 牌号切换次数降低
  • 10% 工厂整体产出提升

重庆博悦机械制造有限公司

矿山破碎机械耐磨件研发与制造 · 装备制造

产品覆盖颚板、轧臼壁、破碎壁、辊皮、齿板等耐磨铸件，长期为破碎设备主机厂及矿业客户配套。

行业挑战：工程图纸是核心技术资料，尺寸公差、材料牌号、热处理与检验标准全部沉淀在图纸中，来源多样、格式涵盖 PDF / JPG。

知识管理痛点：水韧处理、表面粗糙度、铸造公差等级等工艺知识难以结构化沉淀，换型生产、质量检验、客户问询时反复人工查图。

方案应用：以 RatioMax 搭建工程图纸知识库，图纸及技术文档统一入库，多格式自动解析与智能分块；Wiki 模式自动生成结构化知识页面，图纸、零件、材料与工艺概念自动关联。

应用价值 (项目预期目标)

  • 20 份图纸及技术文档统一入库
  • 107 个知识节点构成知识图谱
  • 秒级 自然语言问答替代人工翻图
  • 规范化 质量证据与问题处理

安全与部署

数据合规要求越高，越需要私有化部署

平台支持完整私有化部署，数据完全自主可控。权限、密钥、身份与存储，每一层都为企业级安全纵深而设计。

  • 空间级权限：空间隔离 + 四级角色矩阵的 RBAC 模型，权限可细化到单个知识库，每空间独立审计日志。
  • 两层 API Key：细粒度 API Key（能力级 + 按知识库限制）与平台级 API Key（控制面自动化）分层管理。
  • 单点登录：支持 OIDC 协议接入企业 SSO，沿用现有身份体系。
  • 纵深防护：传输、存储、接口与应用层内置安全加固；MCP 与数据源凭据采用 AES-256-GCM 加密存储，内置 SSRF 防护。

灵活交付，弹性扩缩

  • Docker Compose：单机部署，快速验证
  • Kubernetes：生产级集群部署
  • 微服务架构：四个核心服务独立部署、独立扩缩容，加密通道与消息队列解耦通信
  • 多实例存储：按知识库绑定存储实例：涉密知识库存放私有 OSS，公开知识库存放 S3

生态与集成

嵌入企业已有的工具与基础设施

国产大模型与国际模型自由组合，主流 IM 渠道统一接入，兼容主流云厂商的存储与向量库基础设施——按需组合、随取随用。

IM 渠道

企业微信、飞书、钉钉、云之家、微信、QQBot、Slack、Telegram、Mattermost

模型厂商

DeepSeek、Qwen、智谱、混元、MiniMax、OpenAI、Gemini、Anthropic、Azure OpenAI、NVIDIA、Ollama

向量库

pgvector、Milvus、Qdrant、Weaviate、OpenSearch、Apache Doris、腾讯云 VectorDB

数据源

飞书知识库、飞书云盘、Notion、语雀、RSS、微信小程序、Chrome 插件

MCP 标准工具协议

发布为官方 PyPI 包 ratiomax-mcp，覆盖知识库管理、文档检索、Wiki 操作、会话管理等能力。第三方应用通过标准 OAuth2 获取令牌，敏感操作需人工确认后执行。

  • Chrome 插件：选中文本、图片或整个页面，一键采集为知识条目
  • 微信小程序：轻量移动端，在微信内直接向知识库提问
  • 微信对话开放平台：零代码部署，公众号、小程序即问即答
  • 网站嵌入 Widget：安全模式 Token 交换 + 限流，将智能体发布到外部站点

适用边界

这几类情况，我们可能不是最优解

  • 只需要一个可全文检索的资料库：如果核心诉求是文档存储、分类与关键词搜索，通用网盘或 OA 附件库这类成熟方案通常更轻、更省，也更容易融入现有流程。
  • 只需要一个通用聊天助手：如果日常只是通用问答，并不依赖企业自有知识与引用来源，公有云的通用大模型产品按用量计费，起步成本更低。
  • 暂不具备本地化部署与运维投入：RatioMax 面向愿意自建、对数据自主可控有要求的企业。如果暂时不想承担运维，选择开箱即用的托管型产品会更合适。
  • 只想要结论，不需要可追溯引用：如果答案可以直接采信、不需要展示出处，我们的引用标注与溯源链路就未必是必需项，反而增加使用步骤。
  • 不需要把知识能力接进自有业务系统：如果知识服务只在平台内使用，不打算通过标准化接口嵌入现有系统，MCP / API 的集成能力对你价值有限。
  • 预算与周期尚不足以支撑知识治理：知识治理需要持续的梳理与运营投入。如果短期内难以安排，建议先在一个部门或一类文档做小范围试点，验证后再扩大。

产品能力时间线

持续高频迭代

  • 智能体模式上线
  • 共享空间与技能
  • 全渠道覆盖：企微、飞书、Slack、TG、钉钉
  • 云端与生态
  • Wiki 知识结构化
  • 万级文档与调试
  • 空间级 RBAC 权限
  • 解析双引擎：OpenDataLoader + PaddleOCR-VL
  • 权限与运维
  • 审计与追踪：Langfuse 全链路
  • 文档站与分块治理

关于我们

让企业的每一份文档，都成为可对话、可推理、可复用的智能资产

汇理极知（RatioMax）专注企业级 AI 知识库与智能体平台。产品已在改性工程塑料、矿山机械制造等行业落地，覆盖知识资产化、智能问答、Wiki 结构化、多渠道集成等完整场景。

数据驱动 · 智能未来

常见问题

关于 RatioMax，企业最常问的问题

汇理极知 RatioMax 是什么？
汇理极知（RatioMax）是企业级 AI 知识库与智能体一体化平台。它把散落在网盘、文档与 IM 中的非结构化资料自动解析、分块并向量化，结合 RAG 检索增强生成、ReAct 智能体推理与自动 Wiki 知识图谱，为企业提供有据可依、来源可追溯的智能问答与知识服务。
RatioMax 支持私有化部署吗？
支持完整私有化部署，可运行在企业自有服务器或内网环境，数据完全自主可控。提供 Docker Compose 单机部署与 Kubernetes 生产级集群部署；四个核心服务可独立部署、独立扩缩容，并可按知识库绑定私有 OSS 或 S3 存储实例。
如何保证 AI 回答准确、可追溯？
RatioMax 先检索企业知识再组织回答：语义向量与关键词混合检索，结合 ReRank 智能重排提升命中精度；每个回答自动标注引用来源，点击即可查看原文段落。平台集成 Langfuse 与 W3C 跨服务追踪，从提问入口到知识检索全链路可定位、可回放。
支持哪些大模型？可以使用国产大模型吗？
兼容二十余家模型厂商，包括 DeepSeek、通义千问 Qwen、智谱、腾讯混元、MiniMax 等国产大模型，以及 OpenAI、Gemini、Anthropic、Azure OpenAI、NVIDIA，也可通过 Ollama 接入本地模型，按场景自由组合。
能接入企业微信、飞书、钉钉吗？
可以。RatioMax 支持企业微信、飞书、钉钉、云之家、微信、QQBot、Slack、Telegram、Mattermost 等 IM 渠道，另有微信小程序、Chrome 插件与网站嵌入 Widget，在多个渠道提供一致的知识服务。
支持哪些文档格式和数据源？
可智能解析 PDF、Word、Excel、图片、EPUB、MHTML 等十余种格式：复杂版面 PDF 由 OpenDataLoader 解析，图片与扫描件由 PaddleOCR-VL 识别。支持飞书知识库 / 云盘、Notion、语雀、RSS 自动定时同步，以及 Chrome 插件网页采集与 API / CLI 上传。
RatioMax 如何做知识治理？
提供分块可视化编辑、实时调试面板、版本快照、diff 对比与一键回滚，修改后自动重建索引，并支持自定义元数据检索过滤。Wiki 模式由 Agent 从原始文档自动生成相互链接的知识页面与知识图谱，让知识持续结构化、可维护。
数据安全与权限如何保障？
采用空间隔离与四级角色矩阵的 RBAC 权限模型，可细化到单个知识库，每个空间独立审计日志；支持 OIDC 协议接入企业单点登录；两层 API Key 分级管理；MCP 与数据源凭据采用 AES-256-GCM 加密存储，并内置 SSRF 防护。
什么是 MCP？RatioMax 如何对外提供知识服务？
MCP（Model Context Protocol）是连接 AI 应用与外部工具的标准协议。RatioMax 发布了官方 PyPI 包 ratiomax-mcp，提供 29 个标准工具，支持 stdio / SSE / HTTP 三种传输方式与 OAuth2 授权，敏感操作需人工确认后执行；平台另有约 360 个 API 端点可供集成。
RatioMax 适合哪些企业和行业？
适合数据合规要求高、需要私有化部署的企业（如金融、政务、医疗），以及需要智能体多步推理、知识结构化与多渠道知识服务的组织。目前已在改性工程塑料（广东中塑新材料，股票代码 301686）和矿山机械制造（重庆博悦机械）等行业落地。

联系我们

预约一次专属演示

留下您的需求，我们的解决方案团队将与您联系，结合业务场景评估私有化部署与落地路径。

商务咨询：sales@ratiomax.cn

技术支持：support@ratiomax.cn

RatioMax · Enterprise AI Knowledge Base & Agent Platform | Private Deployment

RatioMax Enterprise AI Knowledge Base & Agent Platform: Bring your documents to life

RatioMax turns unstructured content scattered across drives, docs and IM into knowledge assets your teams can search, reason over and reuse. RAG, agent reasoning and auto-generated Wiki in one platform — every answer cited, every byte under your control.

  • 20+ Model vendors supported
  • 29 Standard MCP tools
  • 7 Vector DB backends
  • 10+ Document formats

Trusted by these enterprises

  • Guangdong Zhongsu New Materials Co., Ltd. · Stock code 301686
  • Chongqing Boyue Machinery Manufacturing Co., Ltd. · Wear parts for mining crushers

Positioning

From static document storage to living knowledge

Traditional document systems only store and search. RatioMax is the intelligence layer on top of enterprise knowledge management — turning static files into dynamic assets that AI can call and business workflows can integrate.

  • Storage: Traditional document systems — Upload, folders, versions; RatioMax · Living knowledge — Multi-source sync, multi-format parsing, structured ingestion
  • Retrieval: Traditional document systems — Keyword full-text search; RatioMax · Living knowledge — Semantic vectors + keywords hybrid search + reranking
  • Structuring: Traditional document systems — None; RatioMax · Living knowledge — Agents autonomously build Wiki pages and knowledge graphs
  • Application: Traditional document systems — Manual lookup; RatioMax · Living knowledge — Multi-step agent reasoning, tool calls, multi-channel service
  • Service: Traditional document systems — No standard interface; RatioMax · Living knowledge — MCP protocol · 29 tools · 3 transports

Core value

Three dimensions that reshape enterprise knowledge

Knowledge as assets

  • Scheduled sync from Feishu Wiki / Drive, Notion, Yuque and RSS; Chrome extension capture; API / CLI upload
  • Parses 10+ formats including PDF, Word, images, Excel, EPUB and MHTML
  • Agents autonomously generate Wiki pages and knowledge graphs with cross-document links
  • Visual chunk editing, version snapshots, diffs, one-click rollback and auto re-indexing

Intelligent services

  • ReAct (reason + act) loops drive multi-step tasks with parallel tool calls and thinking modes
  • Hybrid semantic + keyword retrieval with reranking and cross-base search for higher precision
  • A dedicated data-analysis agent for querying and analysing data
  • Langfuse tracing covers the full path from question to retrieval

Standardized integration

  • MCP interface: 29 tools over stdio / SSE / HTTP with OAuth2 and human approval
  • WeCom, Feishu, DingTalk, Yunzhijia, Slack, Telegram and more IM channels
  • 20+ model vendors including DeepSeek, Qwen, Zhipu, Hunyuan, OpenAI and Gemini
  • Direct file links let third-party apps load files without re-authentication

Data flow

From a single document to a cited answer

Data moves through four stages — each one observable and debuggable.

  1. Ingest: Documents arrive via API, IM or connectors; an ingestion job is created and queued.
  2. Structure: Parsers extract text, chunk it by strategy, then embed it into vector and object storage.
  3. Retrieve: Questions are embedded for semantic search, merged with keyword search and reranked.
  4. Apply: Results enter the model context; the agent reasons, calls tools and answers with citations.

Capabilities

One platform for knowledge and agents

From parsing, chunking and retrieval to reasoning and structuring — every layer is designed for enterprise work.

RAG engine: Retrieve enterprise knowledge first, then answer

Chunking decides RAG quality. RatioMax provides multi-level chunking with full lifecycle management, hybrid retrieval with reranking, and automatic citations — click to see the original paragraph.

  • Multiple parsers: OpenDataLoader for complex PDFs, PaddleOCR-VL for images and scans
  • Live debugging panel to preview chunking parameters instantly
  • Version snapshots, diffs, one-click rollback and custom metadata filters
  • Fan-out across 7 vector DBs: pgvector, Milvus, Qdrant, Weaviate, OpenSearch, Apache Doris, Tencent Cloud VectorDB

Agents & tools: Enterprise agents that reason and act

Agents run ReAct loops to plan steps, call tools, observe results and adapt. Independent tools run in parallel within a single round to cut reasoning turns.

  • Parallel tool calls complete complex tasks in fewer rounds
  • Skill system with preset skills
  • Sandboxed, scoped execution prevents privilege escalation
  • Dedicated data-analysis agent for querying data

Wiki structuring: Turn scattered files into a linked knowledge network

Wiki mode is a RatioMax differentiator. Agents generate structured, interlinked Markdown Wiki pages and knowledge graphs from raw documents — far beyond simple summaries.

  • Extract → generate pages → cross-reference → knowledge graph
  • Page snapshots, line-level diffs, rollback and in-browser editing
  • Task queue with dead-letter queue handles batches of tens of thousands of documents
  • Entities and relations visualised at a glance

Observability: Every answer can be traced, replayed and improved

Langfuse and W3C cross-service tracing record every step from question to retrieval, while task scheduling is built for high concurrency and slow jobs.

  • Staged worker pools keep parsing, embedding and reranking independent
  • Per-model concurrency limits protect model API rate limits
  • Failed jobs land in a dead-letter queue to inspect, retry or discard
  • Prompt cache hit rate and token savings at a glance

Customers

In production across advanced materials and equipment manufacturing

Results shown are project targets

Guangdong Zhongsu New Materials Co., Ltd.

Stock code 301686 · National “Little Giant” specialised enterprise · Engineering plastics

Founded in 2009, focused on R&D, production and sales of modified engineering plastics, with three production bases in Dongguan Chang’an, Ji’an (Jiangxi) and Dongguan Shuixiang.

Industry challenge: High-mix, low-volume, frequent changeovers: weekly schedules took 4–8 hours by hand and re-planning took 1–3 days, relying heavily on planners’ experience.

Knowledge pain point: Formulas, equipment manuals, quality standards and R&D records were scattered across paper and Excel, making changeover lookups slow and error-prone.

Solution: In the Hongmei plant’s AI-powered APS scheduling project, RatioMax delivers the enterprise knowledge base and APS data Q&A, fusing RAG with NL2SQL.

Value (Project target)

  • ≤30 min Planning time (from 4–8 h per run)
  • 80%+ Faster planning-data queries
  • 30–45% Fewer grade changeovers
  • 10% Higher overall plant output

Chongqing Boyue Machinery Manufacturing Co., Ltd.

Wear parts for mining crushers · Equipment manufacturing

Makes wear-resistant castings — jaw plates, concaves, mantles, roll shells and tooth plates — for crusher OEMs and mining customers.

Industry challenge: Engineering drawings hold tolerances, material grades, heat treatment and inspection standards, arriving in mixed PDF / JPG formats.

Knowledge pain point: Process knowledge such as water toughening, surface roughness and casting tolerance grades was hard to capture, forcing repeated manual drawing lookups.

Solution: An engineering-drawing knowledge base on RatioMax with multi-format parsing and smart chunking; Wiki mode auto-generates structured pages linking drawings, parts, materials and processes.

Value (Project target)

  • 20 Drawings & tech docs ingested
  • 107 Knowledge-graph nodes
  • Seconds Natural-language answers vs manual lookup
  • Standard Quality evidence & issue handling

Security & deployment

The stricter your compliance, the more you need private deployment

Deploy the full platform on your own infrastructure with complete data control. Permissions, keys, identity and storage are all built for defense in depth.

  • Space-level permissions: Space isolation with a four-tier role matrix (RBAC), down to individual knowledge bases, with per-space audit logs.
  • Two-tier API keys: Fine-grained keys (capability + knowledge-base scoped) and platform keys for control-plane automation.
  • Single sign-on: Connect your enterprise SSO via OIDC and keep your existing identity system.
  • Defense in depth: Hardened transport, storage, API and app layers; MCP and connector credentials encrypted with AES-256-GCM; SSRF protection built in.

Flexible delivery, elastic scale

  • Docker Compose: Single-node deployment for fast validation
  • Kubernetes: Production-grade cluster deployment
  • Microservices: Four core services deploy and scale independently over encrypted channels and queues
  • Multi-instance storage: Bind storage per knowledge base: classified data on private OSS, public data on S3

Ecosystem

Fits into the tools and infrastructure you already run

Mix Chinese and international models freely, connect mainstream IM channels, and use the storage and vector databases of major cloud providers — compose what you need.

IM channels

WeCom, Feishu, DingTalk, Yunzhijia, WeChat, QQBot, Slack, Telegram, Mattermost

Model vendors

DeepSeek, Qwen, Zhipu, Hunyuan, MiniMax, OpenAI, Gemini, Anthropic, Azure OpenAI, NVIDIA, Ollama

Vector DBs

pgvector, Milvus, Qdrant, Weaviate, OpenSearch, Apache Doris, Tencent VectorDB

Data sources

Feishu Wiki, Feishu Drive, Notion, Yuque, RSS, WeChat Mini Program, Chrome extension

MCP standard tool protocol

Published as the official PyPI package ratiomax-mcp, covering knowledge-base management, retrieval, Wiki operations and sessions. Third-party apps obtain tokens via standard OAuth2, and sensitive actions require human approval.

  • Chrome extension: Capture selected text, images or whole pages as knowledge entries
  • WeChat Mini Program: Ask your knowledge base directly inside WeChat
  • WeChat dialog platform: Zero-code Q&A for Official Accounts and Mini Programs
  • Embeddable widget: Secure token exchange and rate limiting to publish agents on external sites

Where we fit

When we may not be your best fit

  • You mainly need a searchable file repository: If the goal is storing, organising and keyword-searching documents, a general-purpose cloud drive or your OA attachment library is usually lighter and cheaper, and easier to fit into existing workflows.
  • You only need a general-purpose chat assistant: If day-to-day usage is general Q&A without enterprise-specific knowledge or citations, a public-cloud general-purpose LLM product is cheaper to start with.
  • You are not ready for on-premise deployment and operations: RatioMax is built for organisations willing to self-host and keep data under their own control. If you would rather not take on operations, an out-of-the-box managed product will suit you better.
  • You want conclusions only, without traceable citations: If answers can be taken at face value without showing their sources, our citations and traceability may not be essential — they simply add steps.
  • You have no plans to embed knowledge into your own systems: If knowledge is used only inside the platform and never through standard interfaces into your existing systems, our MCP / API integration offers limited value.
  • Budget or timeline cannot yet support knowledge governance: Knowledge governance needs ongoing curation effort. If that is hard to arrange now, start with a small pilot in one team or one document type, then scale once it proves out.

Capability timeline

Continuous, frequent iteration

  • Agent mode launched
  • Shared spaces & skills
  • All channels: WeCom, Feishu, Slack, TG, DingTalk
  • Cloud & ecosystem
  • Wiki knowledge structuring
  • 10k-document scale & debugging
  • Space-level RBAC
  • Dual parsers: OpenDataLoader + PaddleOCR-VL
  • Permissions & operations
  • Audit & tracing: Langfuse end-to-end
  • Docs site & chunk governance

About us

Making every enterprise document something you can talk to, reason with and reuse

RatioMax (汇理极知) builds an enterprise AI knowledge base and agent platform. It is in production in modified engineering plastics and mining machinery, covering knowledge assets, intelligent Q&A, Wiki structuring and multi-channel integration.

Data-driven · Smart future

FAQ

Questions enterprises ask about RatioMax

What is RatioMax?
RatioMax (汇理极知) is an enterprise AI knowledge base and agent platform. It parses, chunks and embeds unstructured content scattered across drives, documents and IM, then combines retrieval-augmented generation (RAG), ReAct agent reasoning and auto-generated Wiki knowledge graphs to deliver grounded answers with traceable sources.
Does RatioMax support private (on-premise) deployment?
Yes. The full platform can run on your own servers or intranet with complete data control. Deploy with Docker Compose on a single node or Kubernetes for production clusters; four core services deploy and scale independently, and each knowledge base can bind to private OSS or S3 storage.
How does RatioMax keep AI answers accurate and traceable?
It retrieves enterprise knowledge before answering, using hybrid semantic and keyword search with reranking. Every answer carries citations that open the original paragraph, and Langfuse plus W3C cross-service tracing record each step from question to retrieval.
Which LLMs are supported? Can we use Chinese models?
RatioMax supports 20+ model vendors, including Chinese models such as DeepSeek, Qwen, Zhipu, Tencent Hunyuan and MiniMax, as well as OpenAI, Gemini, Anthropic, Azure OpenAI and NVIDIA. Local models can be connected through Ollama.
Can it connect to WeCom, Feishu and DingTalk?
Yes. Channels include WeCom, Feishu, DingTalk, Yunzhijia, WeChat, QQBot, Slack, Telegram and Mattermost, plus a WeChat Mini Program, Chrome extension and embeddable website widget.
Which document formats and data sources are supported?
RatioMax parses 10+ formats including PDF, Word, Excel, images, EPUB and MHTML — OpenDataLoader handles complex PDFs and PaddleOCR-VL handles images and scans. It syncs on schedule from Feishu Wiki / Drive, Notion, Yuque and RSS, and accepts Chrome extension capture and API / CLI uploads.
How does RatioMax handle knowledge governance?
Visual chunk editing, a live debugging panel, version snapshots, diffs and one-click rollback with automatic re-indexing, plus custom metadata filters. Wiki mode lets agents generate interlinked pages and knowledge graphs from raw documents so knowledge stays structured and maintainable.
How are data security and permissions handled?
Space isolation with a four-tier RBAC role matrix down to individual knowledge bases, per-space audit logs, OIDC single sign-on, two-tier API keys, AES-256-GCM encryption for MCP and connector credentials, and built-in SSRF protection.
What is MCP and how does RatioMax expose knowledge services?
MCP (Model Context Protocol) is a standard protocol connecting AI applications to external tools. RatioMax publishes the official PyPI package ratiomax-mcp with 29 standard tools over stdio / SSE / HTTP, OAuth2 authorization and human approval for sensitive actions, alongside roughly 360 documented API endpoints.
Which enterprises and industries is RatioMax for?
Organisations with strict data compliance that need private deployment — such as finance, government and healthcare — and teams that need multi-step agents, structured knowledge and multi-channel service. RatioMax is in production in modified engineering plastics (Guangdong Zhongsu New Materials, 301686) and mining machinery (Chongqing Boyue Machinery).

Contact

Book a tailored demo

Tell us what you need. Our solutions team will get in touch to assess private deployment and a practical rollout path for your business.

Sales: sales@ratiomax.cn

Support: support@ratiomax.cn