Is this just AI doing my build for me?
No. The point of Path of Agent is to give average players the same theorycrafting access experienced PoB users already have, not to replace your decisions. The agent operates Path of Building on your behalf and reports results from real PoB calculations. You still pick the direction, you still pick what to actually do. The agent does the testing legwork, not the choosing.
What version of Path of Exile does this work with?
Path of Exile 1. Game data tracks Path of Building and RePoE, so anything those two projects support, this app supports. The League page always shows which league the data currently covers. PoE 2 is not supported.
How does the analysis actually work?
You import a build (PoB code, pastebin URL, or PoE Account character). The desktop app runs Path of Building locally on your PC. We forked PoB and added a programmatic interface with 52 functions: equip an item, swap a gem, allocate a passive, run the calculation, read the result. The agent calls those functions to test changes against your real build state, reads back the resulting stats from PoB, and uses those numbers to pick what to recommend. Every DPS and EHP figure in the output came from a real PoB calculation.
How long does an analysis take?
About 5 to 9 minutes end to end, based on recent runs. The first 2 to 4 minutes are setup: reading your build state, pre-running a batch of test cases, and pulling ladder data for similar characters. After that the agent runs three rounds of testing in Path of Building, and the final summary streams to the screen.
How many tests does an analysis actually run?
Roughly 250 to 400 separate Path of Building calculations per analysis, depending on build complexity. The agent batches them in three rounds and combines the winners into whole-build packages.
What model does it use?
OpenAI GPT-5.5. We tested several smaller and cheaper models. The agent has to chain dozens of tool calls and hold full build context across all of them, and the smaller ones lose track of context, skip verification steps, or invent stat numbers. Until cheaper models close that gap, GPT-5.5 is what we run. Per-analysis cost is expected to fall as cheaper models catch up.
Why credits instead of a subscription?
Each analysis runs the model through dozens of real Path of Building calculations, and we pay OpenAI for every one of them. Credits let you pay only for what you use, no recurring charge. The Oracle's Ledger inside the app shows exactly where every credit went, per call.
How much does an analysis cost?
Around 70 to 90 cents per full analysis at current model prices, or roughly 80 to 100 credits (100 credits ≈ $1 at pack pricing). Follow-up questions are about 15 to 20 cents each. This should drop as cheaper models get good enough to handle the work, and you pay less.
Is there a free trial?
The first 100 beta users get 100 free credits when they make an account and verify their email. That covers your first full analysis and a follow-up question or two. The Build Library is free for everyone, no account required.
What does the server see? What gets stored?
Your build data lives on your PC. To run the analysis the desktop app sends the build context through our server, which forwards it to OpenAI. The server does not keep that data after the analysis finishes, by default. The first time you import a build the app explicitly asks whether you want to share your analysis logs with us for quality improvement and debugging. If you opt in, those logs are kept for 30 days. If you don't, nothing about the analysis is stored on our end. See the privacy policy for the full breakdown.
What platforms?
Windows 10 or newer. The app is built with Tauri.
Why are you asking for feedback?
The product is in beta. Build analysis works and people use it daily, but it is still rough in places and we are figuring out which features matter most. We need real users telling us what they hit, what they want, and what they wish was different. Discord and email are both linked at the bottom of every page.
