@workglow/cactus
v0.3.7
Published
Cactus (needle-rs) tool-calling provider for @workglow/ai.
Readme
@workglow/cactus
needle-rs tool-calling provider for @workglow/ai.
Wraps the 258 KB needle-rs WASM runtime and the 22 MB INT4 SafeTensors Needle 26M model from Abdalrahman/needle-rs-safetensors. Runs in browser (Cache Storage) and Node/Bun (filesystem). Specialized tool-routing only — no free-form chat.
Capabilities
tool-use— function/tool calling viaengine.run/engine.run_streammodel.search,model.info— single-model catalog (Needle 26M)model.download,model.download-remove— fetch + cache the three model files
Entry points
@workglow/cactus/ai— main-thread shell:registerCactus, schema, constants, provider classes@workglow/cactus/ai-runtime— worker server and inline runtime:registerCactusWorker,registerCactusInline, runtime helpers
