Used to analyze Intune error codes with AI. Stored in your browser's localStorage — never sent to any server other than api.anthropic.com.
Model
Pick a model based on cost vs. depth of analysis. Each error-code analysis uses roughly 500 input + 400 output tokens.
Choosing a model
Haiku 4.5 is the default — cheapest, fastest, and has its own rate-limit bucket so it won't conflict with other Anthropic API usage. Sufficient for error-code lookup and most MSI logs where the failure is a single obvious stack trace.
Haiku 4.5 — default. Most error codes and routine logs.
Sonnet 4.6 — escalate when Haiku misses something. Better at correlating timestamps across long IME logs and isolating root cause from noise.
Opus 4.7 — reserve for cases where Sonnet gives up. New tokenizer uses up to 35% more tokens, so effective cost gap is wider than headline pricing.
Biggest cost lever: auto-trim logs before sending — grep for error/return-value lines + surrounding context cuts input tokens 80%+ with no quality loss.