A free, open-source engineering intelligence workbench powered by AI agents
AEC / Engineering · Documents · Media · Code — Built-in Agent | Any API Key | Multi-Agent | Remote Access | Cross-Platform | 24/7 Automation
English | 简体中文 | 繁體中文 | 日本語 | 한국어 | Español | Português | Türkçe | Русский | Українська
PanAI is more than a chat client — it's an engineering intelligence workbench. AI agents run on your own computer with full file access and turn a request into real deliverables: engineering drawings, BIM/IFC models, quantity & cost takeoffs, Office documents, images & video, and code. The agents read your files, run real engines, and you see every action and stay in control.
| Traditional AI chat | PanAI | |
|---|---|---|
| Operate on real project files | Limited or No | Full file access: CAD / BIM / IFC / Office / PDF |
| Produce real engineering deliverables | No | Drawings (DXF/DWG), BIM/IFC, quantity & cost (QTO/BOM), reports |
| Media generation & enhancement | Text / images only | Image + video generation, AI super-resolution (4K), frame interpolation |
| Run multi-step tasks autonomously | Limited | Autonomous, with your approval |
| Remote access from phone | Rarely | WebUI + Telegram / Lark / DingTalk / WeChat |
| Scheduled automation | No | Cron — 24/7 unattended |
| Multiple AI agents at once | No | Built-in agent + Claude Code, Codex, Gemini and more — unified |
| Price | Free / Paid | Free & Open Source (Apache-2.0) |
PanAI's PanAEC assistant turns models and drawings into real, openable files — not just text:
- BIM / IFC — parse, validate (schema + IDS), and derive geometry (in-browser viewer, glTF / 3D Tiles), with property and quantity extraction (ifcopenshell).
- CAD conversion — IFC → STEP / IGES / STL / DXF / DWG / 3DM; import DWG / DXF; parametric CAD generation.
- Quantity & cost (QTO / BOM) — geometry-based takeoff and costing.
- Drawings — plans / elevations / sections exported to DXF + PDF.
Honest about hardware limits: formats that require licensed/x86-only engines (e.g. SketchUp SKP, Revit RVT write-out on ARM) return an explicit "not available here" instead of fabricated output.
- Image generation — style / aspect ratio / resolution up to 4K–8K, with Real-ESRGAN super-resolution for high tiers (cloud provider or local models).
- Video — text-to-video, plus AI super-resolution to 4K (Real-ESRGAN, frame-by-frame on GPU) and RIFE frame interpolation (smoother / higher frame-rate) on existing clips — audio preserved.
PanAI ships with a complete AI agent engine. Unlike tools that require you to install CLI agents separately, PanAI works the moment you install it.
- No CLI tools to install — the agent engine is built in
- No complex setup — paste any API key to get started
- Full agent capabilities — file read/write, web search, image/video generation, MCP (Model Context Protocol) tools
- Ready-to-use assistants — 6 domain assistants (AI开发, 全域中枢, PanAEC, PaNews, Pandolla, 电商运营)
PanAI auto-detects installed CLI agents and runs them through one unified interface, alongside the built-in agent:
Built-in agent · Claude Code · Codex · Qwen Code · Gemini · Hermes Agent · OpenClaw · Nanobot · Cursor Agent · GitHub Copilot · Kimi CLI · OpenCode · Aion CLI (aionrs)
- Auto detection — recognizes installed CLI tools automatically
- Parallel sessions — run multiple agents at once with independent context
- Unified MCP management — configure MCP tools once, sync to all agents
- Full-auto mode — one click to bypass permission prompts for unattended runs
A Leader agent receives your instructions, breaks them into subtasks, and delegates to Teammate agents via a built-in Team MCP server. Teammates run in parallel, share results through an async mailbox, and write to a shared task board — each in its own isolated, approval-gated workspace.
Other AI apps give you a chatbox for your API key. PanAI gives you a full agent — file read/write, web search, image/video generation, tool use — regardless of which model powers it.
| Your API Key | What you get |
|---|---|
| Anthropic | Claude-powered agent |
| OpenAI | GPT-powered agent |
| Gemini | Gemini-powered agent |
| Ollama / LM Studio (local) | Local-model agent |
| Gateway (NewAPI / OpenAI-compatible) | Unified access to many models |
- 6 built-in domain assistants — and you can create your own with custom rules.
- Three-tier skills — built-in (shipped), custom (yours), and extension (third-party) skills, toggled per conversation.
- WebUI — open PanAI in any browser, on phone / tablet / another computer (LAN, cross-network, or server deployment; QR or password login).
- Chat platforms — Telegram · Lark (Feishu) · DingTalk · WeChat.
- Scheduled tasks — cron / interval / one-time triggers for 24/7 unattended runs.
View AI-generated results inline without switching apps: PDF, Word, Excel, PPT, images, video, code, Markdown, HTML, CAD (DXF), BIM (IFC) — with live editing of text/code and Git-based version history.
System requirements: macOS 10.15+ · Windows 10+ · Linux (Ubuntu 18.04+ / Debian 10+ / Fedora 32+) · 4 GB+ RAM.
- Download the installer for your platform from the Releases page.
- Enter any API key to get started.
- Start working — the built-in agent is ready to go.
Released under the Apache-2.0 license.
PanAI is a solo project, built by a single developer with the help of AI coding tools — Claude Code and Codex.
It builds on the open-source AionUi project (Apache-2.0), with thanks to its authors and contributors.