eCan.ai is an AI native, privacy-first, agent app for e-commerce, aimed at empowering sellers to run their multi-channel e-commerce businesses with minimal human overheads. This will be achieved by allowing sellers to run every aspects of their e-commerce business with ai agents, from sourcing/procurement to selling, from marketing research to advertising, from legal counsel to customer service. Hopefully your business can run by agents, so that you can be sitting on the beach :)

With eCan.ai, you can run many agents across many vehicles (a.k.a host computers) on a LAN or WAN or a combo of both. There will be at least 1 host computer that serves as a commander, and other computers serve as vehicles that carries a platoon of agents, a computer or a mobile device such as a smart phone or tablet can also serve as a staff officer, a staff officer computer can monitor and command the agents remotely thru the internet.

eCan.ai adopts the de-facto standard-in-making A2A protocol for messaging between agents. We also supports websocket based real time chat between human and agents and between human and agents over the internet.

eCan.ai has built-in agents for several e-commerce tasks, but you can easily create your own agents, and assign tasks for them to do. tasks can be run on either upon human commands or a pre-assigned one-time or recurring schedule, for example routine daily task can be created to run at certain time each day. Each task will run in its own thread. A task doesn't just run by itself, it requires a skill to run, a skill is a work-flow description. So one must associate a task with an skill for it to run correctly.

eCan.ai builds on top of Langgraph (from Langchain), the state of art agent development framework. Any agent task will require at least one skill, which is work-flow described by langgraph.

eCan.ai includes a customized open-source Flowgram editor to provide a graphic way to author a langgraph based workflow. you can simply drag and drop nodes to the canvas, connect them with edges, and fill in the necessary parameters and/or codes, and be done Once you created a workflow graphically. You can also run/cancel/pause/resume/single-step(on a per-node basis) through the langgraph based workflow. Breakpoints can be added on nodes to pause at certain nodes, and upon pausing, you can inspect the state of the langgraph run, or even modify the state attributes and then resume the run. Flowgram based visual programming of langgraph allows you to focus more effort on work flow rather than coding up the langgraph. Here is a feature list of eCan.ai's skill editor:

• ✅ packaged functional nodes enables building workflow with no code or low code.
• ✅ code editor integrated into code node, can edit code directly on canvas.
• ✅ can run/pause/resume all on canvas.
• ✅ can add/clear breakpoints on nodes to pause at certain nodes.
• ✅ support multi-sheets to make large complicated workflow modular
• ✅ node can be left-right mirrored, or top-down mirrored, to support feedback workflow and ease of edge placement.
• ✅ supports bidirectional translation between flowgram and langgraph. Which means you can load, graphically view, re-layout, a langgraph created by direct python code.

For more detailed user guide of the skill editor, please refer to skill editor.

Browser automation is a critical aspect of e-commerce, eCan.ai has integrated a rich set of state-of-art opensource web automation tools into the app. These includes:

• Browser-use -- easy to use, can be run as a langgraph node callable.
• Crawl4ai -- -- easy to use, can be run as a langgraph node callable.
• Browsebase -- to be added.

While playwright is a state-of-art web automation tool, and pretty much become the de-facto standard for web automation, we do realize it is still not the same as some of the popular fingerprint browser used by e-Commerce sellers, to enable their use, we have added a rich set of tools based on selenium and webdriver (chromedriver) so that you can automate web interface with browsers such as AdsPower

In-browser actions are nice, but they are some times not enough, pop-ups, captchas, etc., sometimes we need computer vision (CV) guided mouse and keyboard actions, we have made a state-of-art OCR service to allow CV guided screen understanding and actions, further allowing you to fully control your browser or any other app in that matter.

We have a rich set of ready to use MCP tools to work with LLM, refer to more details here

We have RAG (Retrieve Augmented Generation) functionality built in to our app, so you can easily put your product related files, policy related files, or any business related files into vector DB and retrieve them to generate response for your customers.

Being AI native means majority of the workflow related data will be feed into LLM, but exposing business data might make one feel uneasy, so we have built in mechanism to protect your privacy:

1. on prem LLM deployment using Ollama and Open Source Models like Qwen3
2. provide RPA tools to allow you to scrape and interact with web (you can save tokens too!)
3. provide extensible filter tools to allow you to filter out sensitive data before it is feed into LLM (can be used in conjunction with option 1.)

You will need to subscribe to "Pro" plan. With "Pro" plan, you can chat with any agent remotely even when you're on the road.

Release Date: December 8, 2025 | Channel: Stable

eCan.ai supports multiple build modes for different use cases:

python build.py fast                    # Fast build with caching and parallel compilation
python build.py fast --force           # Force rebuild (ignore cache)
python build.py fast --skip-frontend   # Skip frontend build
python build.py fast --skip-installer  # Generate executable only

python build.py dev                     # Development build (with console)
python build.py prod                    # Production build (fully optimized)

• Parallel Compilation: Uses multiple CPU cores for faster builds (all modes)
• Parallel Installer Creation: Multi-threaded compression for Windows installers
• Smart Caching: Incremental builds that only recompile changed files (fast mode only)
• Auto Data Collection: Automatically collects data files, binaries, and submodules from key packages
• Cross-platform: Works on Windows, macOS, and Linux
• Optimized Packaging: Excludes unnecessary dependencies to reduce size

• fast: Enables caching and parallel compilation for fastest builds (~2-5 min)
  • Uses ZIP compression for installer (fastest)
  • Minimal data collection for essential packages only
  • Optimized for development speed
• dev: Parallel compilation with console output for debugging (~5-10 min)
  • Uses ZIP compression for installer
  • Full data collection for comprehensive testing
  • Includes debug symbols and console output
• prod: Parallel compilation with full optimization and clean builds (~15-25 min)
  • Uses LZMA compression with solid compression (smallest size)
  • Complete data collection for all packages
  • Maximum compression for distribution

Choose the right mode for your needs:

• Development: Use fast mode for quick iterations (python build.py fast)
• Testing: Use dev mode for comprehensive testing (python build.py dev)
• Distribution: Use prod mode for final releases (python build.py prod)

If build time is critical:

• Skip installer creation: python build.py [mode] --skip-installer
• Use fast mode for development: python build.py fast

Expected installer creation times:

• fast/dev modes: 3-5 minutes (ZIP compression)
• prod mode: 15-25 minutes (LZMA compression with solid compression)

• Executable files are generated in the dist/ directory
• Installation packages (if enabled) are created as something like eCan-1.0.0-windows-amd64-Setup.exe (Windows) or equivalent for other platforms

eCan.ai supports multiple build modes for different use cases:

You can acccess more comprehensive documentations here

Coming soon...

Coming soon...

Visit Our Repo on Github, look at issues and let us know what you like to solve.

• the preferred technical discussion platform is github project discussion panel,
• We also have a Discord server.