OfficePlane

OfficePlane is an open-source agentic runtime for document manipulation. It provides a complete pipeline to ingest, understand, plan changes, execute edits, and verify results across office document formats — all powered by AI agents.

What OfficePlane Does

Capability	Description
Document Ingestion	Upload DOCX/PDF files. Gemini vision extracts a hierarchical structure (chapters, sections, pages) with pgvector embeddings for semantic search.
Plan-Execute-Verify	Describe what you want changed. An AI generates an action tree, executes it atomically, then verifies the result matches your intent.
Content Generation	Generate presentations, reports, and documents from scratch via async agent workflows streamed over SSE.
Task Queue	Redis-backed queue with document-level locking, instant BRPOP dispatch, retry with backoff, and parallel execution across documents.
Agent Teams	Decompose complex tasks across multiple parallel agents with shared task lists, dependency tracking, and pub/sub messaging.
Document Hooks	Attach automated checks to document lifecycle events — compliance reviews, style enforcement, cross-doc sync — triggered on every edit.
Real-Time Updates	WebSocket and SSE streaming for live progress on all operations.

Architecture at a Glance

Every file mutation follows a 5-step pipeline:

┌──────────────┐    ┌──────────────┐    ┌──────────────┐    ┌──────────────┐    ┌──────────────┐
│ 1. Open      │───>│ 2. Queue     │───>│ 3. Change    │───>│ 4. Pretty    │───>│ 5. Get       │
│    Instance  │    │    (atomic)  │    │    (harness) │    │    (format)  │    │    File      │
└──────────────┘    └──────────────┘    └──────────────┘    └──────────────┘    └──────────────┘

1. Open Instance — Acquire an exclusive session on the document
2. Queue — Redis queue serializes all mutations with document-level locking
3. Change — Agent harness reads the file, decides edits, and applies them with real tools
4. Pretty — Format-specific normalization (styles, heading hierarchy, spacing)
5. Get File — Download the result in any supported format

Tech Stack

Layer	Technology
API	FastAPI + Uvicorn (Python)
Database	PostgreSQL 16 + pgvector
Queue / Cache	Redis 7 (task dispatch, locking, pub/sub)
Frontend	Next.js 15 + React + Tailwind CSS
Vision	Google Gemini API
LLM	OpenAI GPT-4o + Gemini fallback
Document Drivers	LibreOffice, Rust native (PyO3), Mock

Quick Links

• Quick Start — Get running in under 5 minutes with Docker
• Architecture — Deep dive into the system design
• Features — Explore each capability in detail
• API Reference — Full endpoint documentation
• Guides — Step-by-step tutorials