Vue.js · GitHub · Frontend

JitOffice Puts Docs, AI, and Approval Flows Behind the Firewall

By 徐小夕 · Oct 30, 2026

Read original on juejin.cn ↗ Google Translate ↗ Alt translation

Most Western AI office suites are SaaS-only, which rules out regulated or secretive enterprises that refuse to send core documents to outside servers. JitOffice is one of the few full-stack suites that offers real private deployment, including self-hosted models and audit trails, so it points to where enterprise AI office tools can go when compliance is the hard constraint.

Summary

JitOffice goes after the slice of the AI office market that public-cloud tools like Tencent Workbuddy, Alibaba Qianwen Office, and ByteDance Doubao Work can't serve: organizations that will not upload contracts, bids, technical docs, or other sensitive data to third-party servers. It bundles an online document editor, spreadsheets, presentations, mind maps, RAG knowledge base, AI drafting, and a configurable approval pipeline into one browser-based platform, with every component deployable on the company's own infrastructure.

The platform layers a web access front end over user, tenant, permission, and approval management, then a capability layer that reuses the self-developed JitWord editing SDK. Models are pluggable via configurable API keys, with support for private and open-source models, RBAC, audit logs, domestic IT adaptation, and integration with existing OA systems. Multi-tenancy isolates each company's data, and a single account can switch between multiple company spaces.

For smaller teams, it collapses separate tools for documents, tables, and approvals into one place. Consulting and delivery teams that serve multiple client companies get clean per-client isolation. The product is scheduled for official launch on November 15, 2026, and is already open for preview requests.

Takeaways

— JitOffice bundles Word, Excel, PPT, and mind maps with AI generation and configurable approval flows in a single web platform.

— The full stack, including document engine, collaboration, AI knowledge base, and assistants, is deployable on enterprise servers or an internal network so data never leaves the domain.

— Public-cloud AI office tools from Tencent, Alibaba, and ByteDance are the motivation: they drove down the barrier to AI adoption but create data-leakage, compliance, and cross-border risks for regulated organizations.

— The platform supports multiple large language models through configurable API keys, including self-built private models and local open-source models.

— Multi-tenancy lets one account create and switch between separate company spaces, with data isolated by a type field and per-tenant storage.

— A configurable approval flow handles submission, review, approval, and rejection with audit records, covering contracts, seal usage, and expense applications.

— The architecture reuses the JitWord editing SDK for all document types, with the platform handling users, spaces, processes, and AI orchestration.

— JitOffice officially launches on November 15, 2026, and preview access is available by request.

— The data flow is: user action, tenant and permission validation, AI or editing SDK processing, then tenant-isolated storage.

Conclusions

The pitch is explicitly anti-SaaS: the product's entire identity is that it does not trust public cloud AI, and that's a positioning the big platforms cannot easily copy without restructuring their own deployment model.

Building on a mature editor SDK instead of reinventing editing keeps the engineering cost of adding new formats low, but it also means the product's differentiator is the orchestration layer, not the editing experience.

A launch date more than a year out suggests this post is part of an early access and partnership push, and the actual product value is unproven until it runs in a real enterprise environment.

Concepts & terms

RAG internal knowledge base

Retrieval-Augmented Generation over an organization's own documents: a model answers queries by first pulling relevant chunks from a private corpus, so responses are grounded in internal material and data stays on-premises.

MCP services

Model Context Protocol services, a standard way to connect AI assistants to external tools and data sources so the model can invoke actions or fetch additional context at runtime.

Xinchuang adaptation

Meeting China's 'Information Technology Application Innovation' requirements, meaning the software runs on domestic CPUs, operating systems, databases, and other infrastructure rather than foreign components.

Source: juejin.cn ↗ Google Translate ↗ Backup ↗