Path of Agent is a desktop companion for Path of Exile 1. Import your build, run a unified analysis that tests gear, skills, and passive-tree changes together inside Path of Building, and get concrete upgrade packages with stat diffs measured — not predicted.
One core analysis engine, plus the reference material and account plumbing to make it usable.
A single orchestrator drives three rounds of PoB testing — individual changes first, then cross-category combinations — until the strongest upgrade packages emerge. Every stat diff comes from PoB, not the model.
Each guide pins near-median ladder characters at Early Mapping, Endgame, and Aspirational tiers, with full gear, gem links, and passive tree for every tier. Plus a leveling section with a scrub-through tree plan.
A six-tab drawer — Overview, Skills, Gear, Tree, Configs, Progression — pulls your archetype's ladder distribution into the app. The Progression tab shows how similar builds scale from early mapping to endgame.
OAuth sign-in with your Path of Exile account. Import any public character by name and the session sticks across launches. No more pasting PoB codes from pastebins.
Every recommendation is computed against your actual build using Path of Building.
For those who want to know what's actually going on behind the streamed results.
Not a chatbot guessing at your build — an agent with custom tools operating a full Path of Building simulator.
When you start an analysis, an AI agent receives your full build state and a set of custom tools wired into Path of Building. It can test gem swaps and see the real DPS change. Simulate gear upgrades with actual mod rolls. Respec tree nodes and check that your resistances still cap.
Inside one unified analysis the orchestrator runs three rounds, each driving dozens of PoB simulations while the agent holds full build context the whole way through. Every change is verified against your actual character — no estimates, no hallucinated numbers.
Each tool call triggers dozens of PoB simulations while maintaining full build context across all of them. Cheaper models lose track mid-chain, skip verification steps, or hallucinate stat values. We tested extensively — only frontier-class models can handle the multi-step reasoning this analysis demands.
That model quality has a real cost per token. Credits let you pay for what you use — no subscriptions, no surprise charges. The Oracle's Ledger inside the app shows exactly where every credit goes.
A full unified analysis runs about $0.50–0.70. Credits never expire.
Screenshots, bug reports, feature ideas — the Discord is where the next features get shaped. Every report helps.
