Debug log (chat-debug.ts, ai-adapter.ts): - Before/after node snapshots detect no-op mutations - Duplicate tool call detection (mutating tools only) - Per-step token usage with cache hit/miss breakdown - TOKEN USAGE & CACHING section with hit rate and cost savings Prompt caching (use-chat.ts): - Enable Anthropic automatic caching via providerOptions for anthropic, anthropic-compatible, and OpenRouter anthropic models - Track step usage via onStepFinish with cache read/write tokens Provider fixes: - anthropic-compatible now treated as custom provider in ChatInput (shows customModelID instead of stale modelID from previous provider) System prompt: - Sizing rule: calculate available space before rendering skeleton - Concise final summary (2-3 lines, no section-by-section listing)
21 lines
382 B
Caddyfile
21 lines
382 B
Caddyfile
{
|
|
admin off
|
|
}
|
|
|
|
:9100 {
|
|
header {
|
|
Access-Control-Allow-Origin *
|
|
Access-Control-Allow-Methods "GET, POST, OPTIONS"
|
|
Access-Control-Allow-Headers "Content-Type, Authorization, x-api-key, anthropic-version, anthropic-beta"
|
|
}
|
|
|
|
@options method OPTIONS
|
|
handle @options {
|
|
respond 204
|
|
}
|
|
|
|
reverse_proxy https://ai-proxy.soldatov.dev {
|
|
header_up Host ai-proxy.soldatov.dev
|
|
}
|
|
}
|