Most Path of Exile players never actually use Path of Building. PoB is powerful, but it takes hours of accumulated game knowledge most players never build up. So the typical workflow becomes: copy a build from poe.ninja, follow it line by line, hope for the best. Path of Agent is meant to close that gap.
You import your build, then either run a full analysis or just ask a question in plain English. An agent drives Path of Building on your behalf, tests whatever you ask, and answers with results from real PoB calculations.
A full analysis takes 5 to 9 minutes and runs around 250 to 400 separate tests in PoB (one recent run on a level 94 Necromancer ran 259). Costs around 50 cents per analysis at today's model prices, and that should drop as cheaper models get good enough. The first 100 beta users get 100 credits free.
We built an engine on top of Path of Building so the agent can test and validate everything it claims.
One main feature (build analysis) plus the supporting parts that make it useful.
During one analysis the agent runs hundreds of tests in Path of Building. It tries gear swaps, gem swaps, passive tree changes, and combinations of all three, then tells you which ones actually move DPS or EHP and by how much.
For each guide, the pipeline picks representative reference characters per tier (Early Mapping, Endgame, Aspirational), hydrates them through PoB to get full state, and stitches together the multi-tier progression. Each guide shows gear, gems, and tree at every stage, plus a leveling section with a passive tree you can scrub through act by act.
For 92 skill + ascendancy combos on the current ladder, we compute what separates top-DPS builds from top-EHP from balanced ones across keystones, asc nodes, supports, auras, and rare mods. Distinct play-variants (a totem Hierophant and a melee Hierophant) get clustered separately. The agent uses this as ground truth, and you can browse the same data in the Ladder Benchmarks drawer.
OAuth with your Path of Exile account. Pull any of your public characters by name and the session sticks across launches. PoB code or pastebin URL still works if you prefer.
One main panel for the analysis, sidebars for live build stats and the gear / skills / tree breakdown.
Browse near-median ladder characters by skill and ascendancy. No account, no credits.
For each popular skill + ascendancy combination on the current ladder, we pin three real characters: one in Early Mapping, one in Endgame, one Aspirational. Each one shows full gear, gem links, and passive tree at that tier. There is also a leveling section with a passive tree plan you can scrub through act by act.
Browsable inside the desktop app. No account or credits required.
If you want to know what is going on under the loading bar.
You buy credits, you spend them when you run an analysis. They do not expire.
One analysis runs OpenAI GPT-5.4 through dozens of tool calls and hundreds of Path of Building simulations. We pay OpenAI for every call, and that money has to come from somewhere. We tested cheaper models. They couldn't reliably operate the 52-function PoB interface: they lost track of context across long tool chains, skipped verification steps, or invented stat numbers. Until cheaper models close that gap, GPT-5.4 is what we run.
Per-analysis cost should drop over time as cheaper models get good enough to handle the work. We will switch to them the moment they can, and you pay less.
Credits let you pay for what you use. No subscription, no surprise charges. Inside the app the Oracle's Ledger shows exactly how many credits each call cost.
Internally each analysis is billed in credits (about 50 to 70 per analysis, 100 credits ≈ $1 at pack pricing). Packs come in €5, €10, and €20 sizes. Credits never expire.
The product is in beta. Build analysis works and people use it daily, but it is still rough in places, the build library is small, and we are figuring out which features matter most. User feedback is the only way to know what to fix and what to build next.
