add: Hosted generation for the Space, and retire the ZeroGPU attempt

fix: Build the Space's agent at import scope, where ZeroGPU is watching

fix: Load the Space's model the one way ZeroGPU allows

fix: Unpin the Space's transformers -- the pin made the build unsolvable

fix: Serve the Space a pre-quantised checkpoint

add: A CUDA path for the demo Space, behind the existing engine contract

refactor: Collapse four inference backends into one MLX path

add: Truncate conversation history to a fixed token budget

test: Try float16 instead of bfloat16 for the PyTorch CUDA path

fix: Stop the PyTorch backend from using pure greedy decoding

add: CONTROLAI_BACKEND=pytorch escape hatch to bypass the GGUF path

fix: Wire up the GGUF deployment path and fix its tokenizer mismatch

fix: Stop final synthesis from hallucinating past a fabrication refusal

fix: Repetition-loop generation, redundant tool spam, and KaTeX placeholder leak

fix: Fall back to end-of-string match for truncated tool-call JSON

fix: Extend provenance guard to 1D arrays; repair trailing-comma JSON

fix: Stop silent empty-answer fallback; surface tool-error reasons

fix: Enforce parameter provenance, fix rendering/citations, add tools, harden agent loop

chore: Remove dead code and stale Docker/GGUF-era artifacts

debug: Add timing/token-count diagnostics and explicit EOS to PyTorch generate()

fix: Explicitly .to('cuda') instead of device_map='auto' on ZeroGPU

perf: Use SDPA attention on CUDA instead of the eager default

fix: Increase llama.cpp context window to fit the system prompt + tool schemas

fix: Load the actual fine-tuned ControlAI model on HF Spaces, not stock Qwen2.5

feat: Enforce high-speed GGUF Q4_K_M llama_cpp execution on Hugging Face Spaces CPU

perf: Preload model on container startup and optimize PyTorch CPU multi-threading

fix: Restore fine-tuned LoRA agent default loading, space-separated array parsing, and plot rendering

fix: Ensure fine-tuned LoRA adapter is loaded by default and fix math placeholder collision in web UI

fix: Protect all LaTeX math, fractions, Greek letters, and matrix environments from markdown parser

fix: Enforce Python tool execution for plots and sanitize raw LaTeX figure tags in Web UI

feat: Add high-speed C++ GGUF and Ollama engine backends to agent orchestrator

fix: Simplify Colab notebook, fix rank_bm25 import handling, and clean up prompts

feat: Add universal PyTorch/Transformers backend for Linux and Google Colab

feat: Initial open-source release of ControlAI