@txt2sfx/agent
v0.1.0
Published
The LLM half of the pipeline: providers, the system prompt built from the live grammar tables, and the generate-validate-render-optimize loop. Zero runtime dependencies — providers speak HTTP through `fetch`.
Readme
@txt2sfx/agent
The LLM half of the txt2sfx pipeline: providers, the system prompt built from the live grammar tables, and the generate → validate → render → fit loop. Providers speak HTTP through fetch — zero runtime dependencies, and no API key of its own: the key comes from the caller.
Takes a render function you provide; no audio stack is bundled. The loop asks the model for a structural change only when the numerical optimizer is stuck — a search that is still improving is never thrown away.
All @txt2sfx/* packages are published in lockstep from one commit — install them at one version.
Apache-2.0.
