An AI enforcement architecture with embedding-based safety routing, multi-model debate, cryptographic provenance chains, and zero-knowledge proofs — built in Python and Rust across 8 repositories.
pip install chimeraforgeWhat powers the platform
Banterpacks is the constitutional AI core. Banterhearts provides the research backbone. The rest of the ecosystem extends access — mobile, web, messaging, observability.
Constitutional AI Core
Constitutional enforcement architecture — embedding-based safety routing (99% fast-path), multi-model debate engine, Rust runtime with Ed25519 provenance and ZK proofs, RLAIF self-improving alignment loop, and the JARVIS AI gateway.
ML Research Backbone
Production research platform with CUDA event timing, quantization pipelines, TensorRT compilation, and capacity planning across 36 technical reports.
Muse Protocol — observability + content
6-agent pipeline with ClickHouse analytics. Also the ecosystem observability layer (OTel, Datadog, DLQ).
On PyPI — pip install chimeraforge
LLM deployment optimizer. 4-gate capacity planner across model, quant, backend, and agent count for 15 GPUs.
Mobile companion
Unity Android client — voice, chat, tool approval, session handoff, mesh networking, offline-first.
Channel adapters
Slack and Discord bridges to JARVIS with session tracking and device key auth.
Web dashboard
Next.js console — chat with streaming, control room, cognitive ELO, tool catalog, workflows.
Automated documentation
Ingests commits from all repos, generates multi-persona episodes, publishes the research archive.
Each episode surfaces commits, telemetry, and engineering decisions as they happen. Four AI personas debate the work — keeping the rough edges in.
feat: TR134 Alignment Under Quantization + TR135/136 Scaffold
This **alignment robustness** episode represents the **Phase 3 threshold**—the moment when Chimera stops asking "does it run?" and starts asking "does it stay aligned?" With 25,258 lines added across 69 files in 6 commits, this update demonstrates **safety-under-quantization m...
docs+style+fix: Final READMEs, Conclusive Reports, Codebase Polish, CI Restoration
This **Phase 2 culmination** episode represents the **completion singularity**—the moment when fourteen episodes of research, benchmarking, profiling, and optimization are distilled into conclusive documentation, unified formatting, and a CI pipeline that actually runs. With 2...
feat: Legacy Cleanup + TR133 Predictive Capacity Planner
This **legacy cleanup + predictive planning** episode represents the **operational singularity**—the moment when Chimera simultaneously retires what no longer serves it and builds the intelligence to plan what comes next. With 5,422 lines added and 598 removed across 84 files...
feat: TR129-TR132 — N-Agent Scaling, Serving-Stack Overhead, GPU Profiling, In-Container Kernel Analysis
This **mega research sprint** episode represents the **investigation singularity** — the moment when Chimera stops asking "how fast?" and starts demanding "why not faster?" With 22,383 lines added across 63 files in just 36 hours, this update demonstrates **relentless empirica...
feat: TR128 Production Workload Characterization
This **production workload characterization** episode represents the **reality singularity**—the moment when Chimera confronts how real traffic behaves on consumer GPU hardware and discovers that **theory diverges from practice**. With 7,432 lines added across 19 files and 3 c...
feat: TR127 Long-Context Performance Characterization
This **long-context performance characterization** episode represents the **scaling truth singularity**—the moment when Chimera stops treating context-length scaling as a single curve and discovers it is two entirely different phenomena. With 5,362 lines added across 15 files,...
Research Program
Independent ML research with CUDA event timing and controlled safety evaluations. Every claim is backed by data, every optimization is measured.
Model loading, ONNX conversion, quantization baselines, security analysis.
Cross-backend inference parity, TensorRT compilation, scaling laws.
KV cache tuning, INT8/FP8 quantization, context scaling, capacity planning.
Alignment under quantization, batch perturbation, multi-turn jailbreaks, cross-architecture fragility, cross-request composition.