Privacy

Accept optional first-party analytics or decline. Functional journey and sound preferences stay on this device.

Read the privacy notice

Open to talks and workshops

SkillOctober 10, 2026

Deterministic LLM Fallback (Claude Skill)

Free Claude Skill, MIT licensed. Ported from the deterministic mode engine in Alex OS on this site.

Keeps an AI feature fast and honest when the model is down, out of credit or switched off, with a circuit breaker and extractive answers.

Treats no-LLM as a product state. A per-provider circuit breaker opens instantly on credit or auth errors and never on a plain 400. The no-LLM path does intent classification, an IDF + MMR deterministic rerank and extractive answers that quote published text with sources, and generates nothing. The same eval suite runs in both modes. Where it comes from: My site's assistant ran on this path for days after an API credit balance ran out, and it still answered in under a second.

Assistants and RAG features that must survive outages, budget caps and key rotation.

# Claude Code (personal, all projects)
unzip deterministic-llm-fallback.zip -d ~/.claude/skills/

# Claude Code (one project)
unzip deterministic-llm-fallback.zip -d .claude/skills/

# Claude.ai / Claude desktop
Settings → Capabilities → Skills → Upload skill → choose the zip
What's inside
  • SKILL.md: audit → breaker → no-LLM path → evals
  • scripts/breaker.py: per-provider circuit breaker (stdlib)
  • scripts/extractive.py: deterministic rerank + quote-only answers (stdlib)
Deterministic LLM Fallback (Claude Skill) · Alexandra Nac