Software that is built, deployed, and testable today.

What it is: A zero-dependency Python verification and memory engine (tb_cs_engine.py) exposed over the Model Context Protocol (2025-03-26 specification) at resonantexo.com. It connects in one command to Claude Code, Cursor, Windsurf, or Google Antigravity.

How it works: Provides seven deterministic tools (annotated with readOnlyHint, destructiveHint, idempotentHint, and openWorldHint: false) that run in milliseconds without calling an LLM: (1) resonant_verify_cegar diffs old vs. new code contracts and runs sandboxed Python bytecode across boundary inputs to catch behavioral regressions, broken signatures, blocked async loops, real-time audio allocations, 60fps UI memoization bugs (UI_SEQ_IN_MEMO_SIG), and CWE shortcuts before save (emitting JSON or OASIS SARIF v2.1.0); (2) resonant_architect_orient extracts only the target function contracts and shared-state callers across files (~87% fewer context tokens); (3) resonant_memory_jtms links project architectural decisions to normalized AST SHA-256 hashes so stale notes are automatically invalidated when code changes; (4) resonant_evolve_guardrail synthesizes reusable AST checks from bug/fix pairs; (5) resonant_honesty_audit verifies completion summaries against live symbol tables and test exit codes; (6) resonant_ci_gate enforces identical pre-merge checks in GitHub Actions / GitLab CI; and (7) resonant_product_telemetry reports FinOps token compression and verification metrics.

3-Tier Access Model: Anyone can test the real Python engine immediately via the free public sandbox at https://resonantexo.com/api/mcp (50 calls/day per IP, isolated sandbox). Evaluators and pilot partners receive signed X-Resonant-Key tokens (5,000 calls/day with persistent isolated SQLite memory graphs), while local/CI installs run unrestricted offline on the developer’s machine.

What it is: The cloud-hosted FastAPI and SQLite backend that powers the remote Resonant MCP server, multi-session contradiction-damped memory graphs, and tenant key verification on Google Cloud Run (backed by Google Vertex AI, the Google Gemini API, Cloud Storage FUSE, BigQuery FinOps telemetry, and 24/7 Cloud Monitoring HTTPS uptime probes), alongside OpenAI-compatible provider adapters for NVIDIA NIM API, Ollama, and vLLM-compatible endpoints.

How it works: Standard vector databases retrieve text chunks by embedding similarity alone, often mixing outdated notes from weeks ago with current specifications. Resonant Cloud stores architectural decisions in a time-stamped graph with 2-hop contradiction damping and AST hash bindings, returning compact ~360-token active context slices that maximize prompt KV-cache hit rates across Google Gemini/Vertex AI, Anthropic Claude, OpenAI, and OpenAI-compatible NIM / vLLM endpoints.

What it is: A browser-playable 3D dungeon crawler, standalone Windows desktop executable, and Android application built on top of four classic open-source C/C++ roguelike engines (Angband, Sil-Q, FrogComposband, and Hengband), live at angband3d.com (v11.1.1).

Why we maintain it as a public proving ground: Maintaining a real-time 60fps Three.js/WebGL2 GPU-instanced renderer bridged to legacy C/C++ WebAssembly workers, zero-flicker content-digest DOM overlays, procedural WebAudio synthesizers, and cloud container backends is the exact kind of tightly coupled, multi-language codebase where unassisted AI coding agents routinely break downstream callers or introduce per-frame memory leaks. Every release of Angband 3D is built and verified under Resonant MCP guardrails.

What it is: A deterministic 64-bit Python and browser-based WebAudio signal-processing workstation built for capturing, mixing, mastering, and publishing live multi-piece acoustic ensemble performances. Starting from a portable 12-track Broadcast Wave (BWF 48,000 Hz / 24-bit) stage recording on a Tascam Model 12 and adapting to ensembles with variable input counts and room acoustics, Studio-12 turns raw multi-mic live board captures into finished albums and hosted digital media kits.

End-to-End Workflow Pipeline: (1) Live Stage Acquisition: Captures discrete 24-bit / 48 kHz multi-track BWF stems directly from the stage mixer across vocal and instrument channels; (2) Automated Mixer WAV Extraction & Phase Alignment: Extracts multi-gigabyte full-set WAV recordings in ~8 seconds, phase-aligns channels against the bass reference, combines dual-mono ADCs (+3.01 dB SNR gain), and trims inter-song stage chatter with zero destructive noise gates; (3) Role-Aware Ensemble Mixing & Cloud Loading: Loads discrete stems into the browser-based Studio-12 Console with per-channel 3-band EQ (with live Bode plot), compression, stereo stage radar, and 3-role vocal seating (Center Lead, Wide Harmony, Vocal Bass Pocket). Because live ensembles trade lead singers from song to song, engineers can dial in the room sound on one track and propagate the mix across a 20+ song set in one click while automatically preserving each song’s lead and backup roles; (4) 64-Bit Mastering & 4-Way Speaker Check: Auditions the raw board capture against the mastered mix across four speaker profiles (Studio Flat, Car Cabin, Phone/Laptop, Mono ∅) and renders 64-bit masters normalized to -14.0 LUFS (ITU-R BS.1770-4, -1.0 dBTP true-peak ceiling); and (5) Website Publishing & Digital Media Kits: Packages streaming audio, stage photography, technical liner notes, and booking/tip-jar modules into a live Concert Vault website deployed on Cloudflare Pages and Google Cloud Storage.

Commercial Roadmap (Pre-Revenue): Designed for deployment both as a turnkey on-location field recording and digital media kit workflow for live ensembles and venues, and as a repeatable cloud-connected production platform (currently in early-access field validation).