v1.0.1: Fix tokenizer loading, prompt formatting, cache wiring, MCP intelligence#11

Merged

devwhodevs merged 17 commits into

mainfrom

hotfix/v1.0.1-tokenizer-and-review-fixes

Mar 25, 2026

Commits on Mar 25, 2026

fix(llm): add shimmytok fallback for GGUF-embedded tokenizers
devwhodevs
committed
fix(llm): apply prompt format in CandleEmbed embed_one and embed_batch
devwhodevs
committed
fix(store): clear FTS on reindex and use stored dim for vec table init
devwhodevs
committed
fix(search): wire LLM cache into search_with_intelligence
devwhodevs
committed
fix(serve): wire orchestrator and reranker into MCP search handler
devwhodevs
committed
feat(llm): add BERT GGUF architecture support, switch default to all-MiniLM-L6-v2
devwhodevs
committed
feat: add accelerate feature flag for optimized CPU on macOS
devwhodevs
committed
fix: add indexing progress output, fix Qwen3 GGUF filename case
devwhodevs
committed
fix: use float32 RmsNorm for Metal GPU compatibility in Gemma embedding
devwhodevs
committed
refactor(llm): replace candle backend with llama-cpp-2 for Metal GPU support
devwhodevs
committed
feat(llm): switch to llama.cpp backend, fix embedding params
devwhodevs
committed
style: cargo fmt
devwhodevs
committed
ci: install CMake on Ubuntu for llama.cpp build
devwhodevs
committed
feat(search): wire intelligence models into CLI search path
devwhodevs
committed
fix: singleton LlamaBackend and built-in tokenizer for orchestrator/reranker
devwhodevs
committed
fix(llm): global backend singleton, built-in tokenizers, wire CLI intelligence
devwhodevs
committed
docs: update README, CHANGELOG, CLAUDE.md for llama.cpp backend
devwhodevs
committed