TransportX Agent

TransportX Agent — an AI agent working alongside you to solve real-world transportation problems.

Ask about roads, transit, traffic demand, and mobility data. TransportX turns the work into maps, charts, cited findings, and reports while keeping the source files and tool history close at hand.

English · 简体中文

Why TransportX

Coding agents such as Codex and Claude Code have improved quickly, but using them still often means working with code, terminals, and unfamiliar file structures. That remains a real barrier for many domain specialists, particularly non-programmers who need to analyze traffic data or build map-based explanations.

TransportX began with a simple goal: make agent-assisted analysis practical for transport professionals. It is a lightweight desktop workspace for asking questions, inspecting maps and charts, and delivering results with their evidence attached. The core stays small. Modules add the data, knowledge, methods, and tools required by a field, which also makes the platform useful beyond transport.

_{34-second animated preview covering the full story at 2× speed · Click to watch the 68-second video with sound (1080p)}

The video follows the core workflow: start a traffic task, ask a question, analyze it on a shared map, and deliver a report.

Built for traffic analysis

TransportX gives analysts one place to ask questions, inspect evidence, work with maps, and deliver results. Each task has its own workspace, conversation history, files, and resolved capability versions.

A composable module system

TransportX extends through Modules instead of hard-coding domain logic into the app. Modules can wrap Agent Skills from different tools when they use a SKILL.md entry point, then package them with Extensions, datasets, knowledge collections, report templates, and native runtimes. Each task loads only the Modules it needs and records their exact versions, so teams can add another domain without forking the platform and later reopen the task with the same capability set.

Shared GIS context

The agent and analyst work with the same map state. The agent can publish GeoJSON and update layers; the analyst can select features, mark a point, draw a rectangle, or submit the current viewport. TransportX attaches that input to the next message as structured Geo Context, so follow-up analysis keeps the same map revision and spatial scope without asking the analyst to describe the location again.

Product screenshots

The screenshots show how TransportX combines maps, analysis, figures, and reports in one workspace. They capture example tasks; the underlying traffic datasets are not included in this repository.

Monthly passenger-flow report	Event traffic around Shanghai Stadium
Review a generated report with charts and citations, then export it as PDF.	Compare road speeds and parking demand around an event venue on the map.

Metro and shared-bike connections	Holiday ride-hailing demand
Inspect station rankings and nearby transit lines on the same map.	Compare ride-hailing demand before, during, and after a holiday alongside rail arrivals.

Set up an analysis task	Select an area on the map
Choose a model and Module versions when creating a task.	Send a selected map area back to the agent as geographic context.

How it works

The desktop, Web, and CLI clients share one local Agent Host. It manages sessions, resources, models, and versioned Modules; Pi runs the agent loop and calls the selected model, Python analysis environment, and domain services.

Design principles

• The map is both output and input. The agent can create spatial layers, and analysts can return a point, feature, rectangle, or viewport as structured context.
• Work stays inspectable. Sessions retain tool activity, source references, attachments, maps, and derived artifacts instead of flattening everything into chat text.
• Capabilities are explicit. Data, domain knowledge, analysis methods, and native tools are installed as Modules, selected when a task starts, and pinned for recovery.

Quick start

Run the desktop app from source:

npm install
npm run desktop:dev

Before creating the first task, add a Pi-compatible model from New Traffic Task or Settings.

Traffic databases, source knowledge collections, model credentials, and user-installed Modules are not distributed with the repository.

Platform support

The desktop app is built with Electron and React. A local Node.js Agent Host owns sessions and resource boundaries, while Pi runs the model and tool loop. Python 3.10 ships with the desktop application; the video runtime is installed separately with Video Capability.

Documentation

Acknowledgements

TransportX builds on and has benefited from these open-source projects:

• Pi provides the agent runtime, model integration, and tool loop.
• Pi Tau Web Server is the upstream foundation for the browser workspace and Pi RPC session architecture.
• MapLibre GL JS powers interactive map rendering.
• Electron and React provide the desktop and interface foundations.

Thank you to their maintainers and contributors.

Thanks also to these product testers:

@aowang-ai	@yonghenggudu	@runningjian-ui	@tataxing123
@gjs396891501-cyber

License

TransportX Agent is licensed under AGPL-3.0-only. Commercial use is permitted under its terms. For alternative commercial terms covering project-owned code, contact the maintainers.

Upstream MIT-licensed material retains its original terms and notices; see Third-party source notices.