nezumi/9router
1
0
Fork 0
mirror of https://github.com/Nezumi-2711/9router.git synced 2026-09-22 13:38:31 +00:00
Code Issues Packages Projects Releases Wiki Activity

feat(usage): track cached tokens + correct input/output/cache cost (#2209)

Browse Source 
Normalize every provider to one cache-inclusive convention via
canonicalizeUsage() before persist, and price cached + cache_creation as
subsets of prompt_tokens in calculateCostFromTokens() to stop
double-counting. usageRepo now delegates cost math to a single source.
Surface Cached tokens/cost across dashboard (overview, tokens, cost,
details). Merge Claude message_start cache with message_delta output so
cache counts survive. Compatible LLM nodes now allow multiple API-key
connections (key pool).

Co-authored-by: Cursor <cursoragent@cursor.com>
This commit is contained in:
hodtien
2026-07-03 15:18:27 +07:00
committed by decolua
co-authored by Cursor
parent 960f8a0379
commit 54e3245ace
17 changed files with 558 additions and 71 deletions
Download Patch File Download Diff File Expand all files Collapse all files
open-sse/executors/kiro.js
+11
View File
@@ -391,6 +391,11 @@ export class KiroExecutor extends BaseExecutor {
if (metrics && typeof metrics === 'object') {
const inputTokens = metrics.inputTokens || 0;
const outputTokens = metrics.outputTokens || 0;
// ponytail: Amazon Q upstream does not expose cache fields today,
// but pick up cache_read_input_tokens / cache_creation_input_tokens
// if the event shape grows them so cost tracking stays accurate.
const cachedTokens = metrics.cacheReadInputTokens || metrics.cache_read_input_tokens || 0;
const cacheCreationInputTokens = metrics.cacheCreationInputTokens || metrics.cache_creation_input_tokens || 0;
if (inputTokens > 0 || outputTokens > 0) {
state.usage = {
@@ -398,6 +403,12 @@ export class KiroExecutor extends BaseExecutor {
completion_tokens: outputTokens,
total_tokens: inputTokens + outputTokens
};
// Kiro is Claude-backed: inputTokens EXCLUDES cache (Claude convention),
// not inclusive like OpenAI's cached_tokens. Emit cache_read_input_tokens
// (not cached_tokens) so canonicalizeUsage takes the Claude fold path and
// correctly adds cache back into prompt_tokens instead of undercharging.
if (cachedTokens > 0) state.usage.cache_read_input_tokens = cachedTokens;
if (cacheCreationInputTokens > 0) state.usage.cache_creation_input_tokens = cacheCreationInputTokens;
}
}
}