@younndai/lyt-llm
PackagesThe LLM gateway. It routes AI-assisted work through four adapter families, with declared constraints and a per-run budget hard stop.
@younndai/lyt-llm ships inside the lyt CLI — most people get it there. To use it on its own, add the package:
npm install @younndai/lyt-llmThe LLM gateway
lyt-llm gives automators one model surface. A request carries source preferences and hard constraints; routing selects an eligible adapter instead of hiding the decision in an individual vault operation.
Four sources, one surface
The adapter families are Ollama (local HTTP), harness (a local Claude Code or Codex harness invocation), BYOK (your provider account), and ai-relay (managed provider routing). BYOK and ai-relay make external network requests; Ollama is local, while harness behavior is governed by the configured local agent runtime. Constraints can require local-only routing or prohibit external routing for private memscope.
Cost budgets you control
Every run has a per-run cost budget that stops further work when exceeded; monthly budgets are a warning surface. The package registers generation and embedding operations with lyt-runner and ships inside the lyt CLI.