What's New

A new billing page shows refund statistics and a recent activity table — failed AI calls are refunded automatically, with refund count, credits returned, and every charge, deduct, and refund row live from the billing and usage endpoints.

A dedicated community hub announces the first 48-hour Brody AI Game Jam — sample themes, rules in brief, and the judging rubric — plus a built-with-Brody showcase that opens with the first jam.

Type @ in the editor chat to search the project index by class or file name; pick a match to insert an @ClassName token the model resolves through mention resolution.

Press Tab to complete from recent tool names and their argument templates — faster command entry without guessing tool names.

The tool list the model sees is now grouped by category — VFX, animation, UI, testing — cutting token overhead and making tool selection more accurate.

The verify loop now captures play-mode screenshots before and after a change, so the model sees "visual: CHANGED/UNCHANGED" and can confirm edits actually changed the scene.

Tasks survive Unity domain reloads end-to-end, and FPS template tuning no longer misfires during verification.

Landing, portal, pricing, auth, and the Unity editor UI rebuilt with a dark OLED system: glassmorphism cards, emerald/teal gradient accents, ambient glows, and staggered motion.

17 new game, level, and code tools landed in one batch. Highlights: level builder, dungeon layout, enemy AI planner, gameplay verify, navmesh, lighting, terrain, prefab, scene snapshots/rollback, code analysis, and package manager.

Destructive operations are gated behind explicit approval — auto-approve mode keeps fully-automated runs safe.

Live backend health indicator in the dashboard navigation, so you always know the gateway is reachable.

Consecutive read-only tools now execute in parallel (Task.WhenAll) while writes stay sequential — multi-file work finishes in a fraction of the time.

Tool-only turns stay silent and the model summarizes naturally; progress shows in-chat instead of status spam.

In auto mode the Execute button is hidden (Cancel only) — destructive tools are always confirmable, and rate limiting only kicks in below a healthy credit balance.

Credit grants update the visible dashboard balance immediately; launcher port checks were also rewritten to be idempotent.

Register in seconds, get a 7-day free trial with $5 in AI credits (50,000), no credit card required.

GO unlocks 50,000 monthly credits and standard models; Pro adds frontier models (Claude, Gemini, Nemotron) with 100,000 credits. PayPal checkout, verified end-to-end.

A three.js wireframe AI core with an orbiting Unity cube field and starfield replaces the static landing hero.

Live-verified paid models (GPT-4o-mini, Llama 3.3 70B) now lead the fallback pool, with real Unity code output tested over HTTP.

Specialized sub-agents spawn on demand and operate on main-thread-safe orchestration, proven by a headless 10-agent suite.

New analyze_screenshot tool with real AI image understanding, plus verified content generation for audio, image, and 3D.

A fourth execution mode forces persistent-loop sub-agents for fully autonomous, no-approval workflows.

Postgres (Docker-only), native gateway, and dashboard boot from a single launcher with self-healing, idempotent port checks.