mohdel 0.104.0 → 0.104.2

package/js/core/events.js

@@ -62,14 +62,12 @@
  * @property {number} outputTokens
  * @property {number} thinkingTokens
  * @property {number} [cacheWriteInputTokens]
- *   Tokens written to a fresh prompt cache breakpoint, billed at
- *   `cacheWritePrice` (typically 1.25× input on Anthropic). Absent on
- *   providers that don't surface this counter (OpenAI doesn't separately
- *   bill cache writes).
+ *   Input tokens written to a fresh prompt cache breakpoint, billed at
+ *   `cacheWritePrice`. Absent when the provider has no separate
+ *   cache-write counter.
  * @property {number} [cacheReadInputTokens]
- *   Tokens served from prompt cache, billed at `cacheReadPrice` (typically
- *   0.1× input). Set by Anthropic directly and by OpenAI-shape adapters
- *   after subset→additive normalization of `prompt_tokens_details.cached_tokens`.
+ *   Input tokens served from prompt cache, billed at `cacheReadPrice`.
+ *   Absent when the provider has no prompt caching.
  * @property {number} cost
  *   USD, computed from curated pricing. Single number (not a breakdown).
  * @property {Timestamps} timestamps

package/js/session/adapters/_cancelled.js

@@ -22,10 +22,16 @@ import { catalogKey } from '#core/model-id.js'
  * @param {string} output
  * @param {number} inputTokens
  * @param {number} outputTokens
+ * @param {{cacheWriteInputTokens?: number, cacheReadInputTokens?: number}} [extra]
+ *   Optional cache token counts captured before cancellation. Threaded through
+ *   so the cancellation-cost calculation prices any cache writes/reads that
+ *   already happened before the abort.
  * @returns {import('#core/events.js').DoneEvent}
  */
-export function cancelledDone (start, first, envelope, output, inputTokens, outputTokens) {
+export function cancelledDone (start, first, envelope, output, inputTokens, outputTokens, extra = {}) {
   const end = String(process.hrtime.bigint())
+  const cacheWriteInputTokens = extra.cacheWriteInputTokens || 0
+  const cacheReadInputTokens = extra.cacheReadInputTokens || 0
   return {
     type: 'done',
     result: {
@@ -34,9 +40,11 @@ export function cancelledDone (start, first, envelope, output, inputTokens, outp
       inputTokens,
       outputTokens,
       thinkingTokens: 0,
+      ...(cacheWriteInputTokens > 0 && { cacheWriteInputTokens }),
+      ...(cacheReadInputTokens > 0 && { cacheReadInputTokens }),
       cost: costFor(
         catalogKey(envelope.model),
-        { inputTokens, outputTokens, thinkingTokens: 0 }
+        { inputTokens, outputTokens, thinkingTokens: 0, cacheWriteInputTokens, cacheReadInputTokens }
       ),
       timestamps: { start, first: first ?? end, end },
       warning: WARNING_CANCELLED

package/js/session/adapters/_pricing.js

@@ -29,14 +29,10 @@ import { getSpec, setCatalog } from './_catalog.js'
  *   - `cacheWritePrice` → `inputPrice` (graceful for non-caching providers)
  *   - `cacheReadPrice` → `inputPrice`
  *
- * Token-counting conventions:
- *   - Anthropic reports `cache_creation_input_tokens` and `cache_read_input_tokens`
- *     as ADDITIONAL to `input_tokens` (separately billable). The adapter
- *     surfaces them as `cacheWriteInputTokens` / `cacheReadInputTokens`
- *     (write/read pair, matching catalog `cacheWritePrice`/`cacheReadPrice`).
- *   - OpenAI reports `prompt_tokens_details.cached_tokens` as a SUBSET of
- *     `prompt_tokens` (already counted). Adapters subtract before passing
- *     `inputTokens` to keep this function additive across providers.
+ * Token-counting convention: this function is purely additive across
+ * `inputTokens`, `cacheWriteInputTokens`, `cacheReadInputTokens`,
+ * `outputTokens`, and `thinkingTokens`. Adapters normalize provider-specific
+ * shapes (e.g. subset-of-input vs. additional-to-input) before calling here.
  *
  * @param {any} spec  Catalog entry, or `undefined`.
  * @param {{inputTokens?: number, outputTokens?: number, thinkingTokens?: number,

package/js/session/adapters/anthropic.js

@@ -129,7 +129,7 @@ export async function * anthropic (envelope, deps = {}) {
 
     for await (const event of stream) {
       if (signal?.aborted) {
-        yield cancelledDone(start, first, envelope, currentOutput(), inputTokens, outputTokens)
+        yield cancelledDone(start, first, envelope, currentOutput(), inputTokens, outputTokens, { cacheWriteInputTokens: cacheWriteTokens, cacheReadInputTokens: cacheReadTokens })
         return
       }
       switch (event.type) {
@@ -197,7 +197,7 @@ export async function * anthropic (envelope, deps = {}) {
     }
   } catch (e) {
     if (signal?.aborted) {
-      yield cancelledDone(start, first, envelope, currentOutput(), inputTokens, outputTokens)
+      yield cancelledDone(start, first, envelope, currentOutput(), inputTokens, outputTokens, { cacheWriteInputTokens: cacheWriteTokens, cacheReadInputTokens: cacheReadTokens })
       return
     }
     log?.warn({ err: e }, '[mohdel:anthropic] stream failed')
@@ -206,7 +206,7 @@ export async function * anthropic (envelope, deps = {}) {
   }
 
   if (signal?.aborted) {
-    yield cancelledDone(start, first, envelope, currentOutput(), inputTokens, outputTokens)
+    yield cancelledDone(start, first, envelope, currentOutput(), inputTokens, outputTokens, { cacheWriteInputTokens: cacheWriteTokens, cacheReadInputTokens: cacheReadTokens })
     return
   }
 

package/js/session/run.js

@@ -376,16 +376,19 @@ function finalizeSpanOk (span, result, sawDelta = false, maxInterFrameMs = 0) {
  * @param {number} startedAt
  */
 function summarizeDone (result, startedAt) {
-  return {
+  const summary = {
     status: result?.status,
     in: result?.inputTokens || 0,
     out: result?.outputTokens || 0,
-    think: result?.thinkingTokens || 0,
-    cost: result?.cost,
-    warning: result?.warning,
-    totalMs: Date.now() - startedAt,
-    maxInterFrameMs: result?.maxInterFrameMs
+    think: result?.thinkingTokens || 0
   }
+  if (result?.cacheWriteInputTokens) summary.cacheW = result.cacheWriteInputTokens
+  if (result?.cacheReadInputTokens) summary.cacheR = result.cacheReadInputTokens
+  summary.cost = result?.cost
+  if (result?.warning) summary.warning = result.warning
+  summary.totalMs = Date.now() - startedAt
+  if (result?.maxInterFrameMs != null) summary.maxInterFrameMs = result.maxInterFrameMs
+  return summary
 }
 
 /**

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "mohdel",
-  "version": "0.104.0",
+  "version": "0.104.2",
   "license": "MIT",
   "author": {
     "name": "Christophe Le Bars",
@@ -87,7 +87,7 @@
     "@opentelemetry/exporter-trace-otlp-grpc": "^0.217.0",
     "@opentelemetry/sdk-node": "^0.217.0",
     "chalk": "^5.4.0",
-    "mohdel-thin-gate-linux-x64-gnu": "0.104.0"
+    "mohdel-thin-gate-linux-x64-gnu": "0.104.2"
   },
   "dependencies": {
     "@anthropic-ai/sdk": "^0.95.1",