@ramarivera/coding-agent-langfuse 0.1.39 → 0.1.41

package/README.md

@@ -3,9 +3,9 @@
 Universal coding-agent Langfuse backfiller and OTLP exporter helpers.
 
 It imports local histories from Codex, Claude Code, Grok, OpenCode, and Pi into
-Langfuse as session traces with child observations. LLM usage records are emitted
-as Langfuse generations with model names, usage details, and recorded or
-estimated cost details. Tool calls remain child spans under the same session.
+Langfuse as session traces with child observations. LLM usage records are kept
+as observation metadata so historical imports do not create Langfuse billing or
+cost rows. Tool calls remain child spans under the same session.
 
 ```sh
 coding-agent-langfuse-backfill --agents codex,claude,grok,pi,opencode
@@ -112,7 +112,7 @@ npm run test:e2e
 The e2e suite verifies:
 
 - Codex full session traces with messages, reasoning, tool calls, tool results,
-  usage, and costs
+  and usage metadata
 - Follow mode picking up newly written Codex events
 - One CLI run posting reconstructable traces for Claude Code, Codex, Grok,
   OpenCode, and Pi

package/dist/backfill.d.ts

@@ -63,7 +63,9 @@ type FollowSummary = RunSummary & {
 declare const allAgents: AgentName[];
 declare const defaultEndpoint = "https://langfuse.ai.roxasroot.net/otel/v1/traces";
 declare function parseArgs(argv: string[]): BackfillOptions;
-declare function codexEvents(homeDir: string): BackfillEvent[];
+declare function codexEvents(homeDir: string, options?: {
+    sinceMs?: number;
+}): BackfillEvent[];
 declare function claudeEvents(homeDir: string): BackfillEvent[];
 declare function piEvents(homeDir: string): BackfillEvent[];
 declare function grokEvents(homeDir: string): BackfillEvent[];

package/dist/backfill.js

@@ -235,7 +235,7 @@ function listFiles(root, predicate) {
             }
             if (stat.isDirectory())
                 stack.push(path);
-            else if (stat.isFile() && predicate(path))
+            else if (stat.isFile() && predicate(path, stat))
                 out.push(path);
         }
     }
@@ -374,59 +374,32 @@ function usageDetails(usage) {
         details.total = usage.total;
     return Object.keys(details).length > 0 ? details : undefined;
 }
-function pricingForModel(model) {
-    if (!model)
-        return undefined;
-    const normalized = normalizeModelName(model) ?? model;
-    if (normalized === "kimi-for-coding") {
-        return { input: 0.95, output: 4.0, cacheRead: 0.16, cacheWrite: 0 };
-    }
-    if (normalized.includes("accounts/fireworks/routers/kimi-k2p6-turbo")) {
-        return { input: 2.0, output: 8.0, cacheRead: 0.30, cacheWrite: 0 };
-    }
-    if (normalized.includes("accounts/fireworks/models/deepseek-v4-pro")) {
-        return { input: 1.74, output: 3.48, cacheRead: 0.15, cacheWrite: 0 };
-    }
-    if (normalized.includes("DeepSeek-V4-Pro")) {
-        return { input: 2.1, output: 4.4, cacheRead: 0.2, cacheWrite: 0 };
-    }
-    if (normalized.includes("Kimi-K2.6")) {
-        return { input: 1.2, output: 4.5, cacheRead: 0.2, cacheWrite: 0 };
-    }
-    if (normalized.includes("MiniMax-M2.7")) {
-        return { input: 0.3, output: 1.2, cacheRead: 0.06, cacheWrite: 0 };
-    }
-    return undefined;
-}
-function costDetails(usage, model) {
-    if (!usage)
-        return undefined;
-    const rates = pricingForModel(model);
-    if (rates) {
-        const cachedInput = usage.cacheRead ?? 0;
-        const cacheWriteTokens = usage.cacheWrite ?? 0;
-        const regularInput = Math.max((usage.input ?? 0) - cachedInput - cacheWriteTokens, 0);
-        const input = (regularInput * rates.input) / 1_000_000;
-        const output = (((usage.output ?? 0) + (usage.reasoning ?? 0)) * rates.output) /
-            1_000_000;
-        const cache_read = (cachedInput * rates.cacheRead) / 1_000_000;
-        const cache_write = (cacheWriteTokens * rates.cacheWrite) / 1_000_000;
-        const total = input + output + cache_read + cache_write;
-        if (total > 0) {
-            return { input, output, cache_read, cache_write, total, source: "estimated" };
-        }
-    }
-    if (usage.cost !== undefined && usage.cost > 0) {
-        return { total: usage.cost, source: "recorded" };
-    }
-    return undefined;
-}
 function isGenerationEvent(event) {
     return event.usage !== undefined && event.role !== "user" &&
         event.role !== "developer" && event.role !== "system";
 }
-function codexEvents(homeDir) {
-    const files = listFiles(join(homeDir, ".codex/sessions"), (path) => path.endsWith(".jsonl"));
+function codexSessionPathTimeMs(path) {
+    const rolloutMatch = path.match(/rollout-(\d{4})-(\d{2})-(\d{2})T(\d{2})-(\d{2})-(\d{2})/);
+    if (rolloutMatch) {
+        const [, year, month, day, hour, minute, second] = rolloutMatch;
+        return Date.UTC(Number(year), Number(month) - 1, Number(day), Number(hour), Number(minute), Number(second));
+    }
+    const folderMatch = path.match(/[\\/]sessions[\\/](\d{4})[\\/](\d{2})[\\/](\d{2})[\\/]/);
+    if (!folderMatch)
+        return undefined;
+    const [, year, month, day] = folderMatch;
+    return Date.UTC(Number(year), Number(month) - 1, Number(day));
+}
+function fileMayContainCodexWindow(path, stat, options) {
+    if (options.sinceMs === undefined)
+        return true;
+    if (stat.mtimeMs >= options.sinceMs)
+        return true;
+    const pathTimeMs = codexSessionPathTimeMs(path);
+    return pathTimeMs === undefined || pathTimeMs >= options.sinceMs;
+}
+function codexEvents(homeDir, options = {}) {
+    const files = listFiles(join(homeDir, ".codex/sessions"), (path, stat) => path.endsWith(".jsonl") && fileMayContainCodexWindow(path, stat, options));
     const seenTokenCounts = new Set();
     return files.flatMap((path) => {
         const rows = parseJsonl(path).map(asRecord);
@@ -1271,17 +1244,7 @@ function toOtlp(events, options = {}) {
             const modelName = normalizeModelName(event.model);
             const generation = isGenerationEvent(event);
             const usage = usageDetails(event.usage);
-            const cost = costDetails(event.usage, modelName);
             const eventProject = projectMetadata(event.cwd);
-            const costForLangfuse = cost === undefined
-                ? undefined
-                : {
-                    ...(cost.input !== undefined ? { input: cost.input } : {}),
-                    ...(cost.output !== undefined ? { output: cost.output } : {}),
-                    ...(cost.cache_read !== undefined ? { cache_read: cost.cache_read } : {}),
-                    ...(cost.cache_write !== undefined ? { cache_write: cost.cache_write } : {}),
-                    total: cost.total,
-                };
             const attributes = [
                 attr("service.name", `agent.${event.agent}`),
                 attr("deployment.environment", "local"),
@@ -1290,13 +1253,6 @@ function toOtlp(events, options = {}) {
                 attr("session.id", event.sessionId),
                 attr("langfuse.observation.type", generation ? "generation" : "span"),
                 attr("langfuse.observation.model.name", generation ? modelName : undefined),
-                attr("langfuse.observation.usage_details", usage),
-                attr("langfuse.observation.cost_details", costForLangfuse),
-                attr("gen_ai.response.model", generation ? modelName : undefined),
-                attr("gen_ai.usage.input_tokens", usage?.input),
-                attr("gen_ai.usage.output_tokens", usage?.output),
-                attr("gen_ai.usage.total_tokens", usage?.total),
-                attr("gen_ai.usage.cost", cost?.total),
                 attr("agent.name", event.agent),
                 attr("host.name", currentHost),
                 attr("agent.session_id", event.sessionId),
@@ -1315,7 +1271,8 @@ function toOtlp(events, options = {}) {
                 attr("langfuse.observation.metadata.project_folder", eventProject.projectFolder),
                 attr("langfuse.observation.metadata.model", modelName ?? event.model),
                 attr("langfuse.observation.metadata.provider", event.provider),
-                attr("langfuse.observation.metadata.cost_source", cost?.source),
+                attr("langfuse.observation.metadata.usage_details", usage),
+                attr("langfuse.observation.metadata.recorded_cost", event.usage?.cost),
                 attr("langfuse.observation.input", event.input),
                 attr("langfuse.observation.output", event.output),
                 attr("source.path", event.sourcePath),
@@ -1451,7 +1408,7 @@ function describeError(error) {
 function discoverEvents(options) {
     const providers = {
         claude: (inner) => claudeEvents(inner.homeDir),
-        codex: (inner) => codexEvents(inner.homeDir),
+        codex: (inner) => codexEvents(inner.homeDir, { sinceMs: inner.sinceMs }),
         grok: (inner) => grokEvents(inner.homeDir),
         opencode: (inner) => opencodeEvents(inner.homeDir, {
             rowLimit: inner.limit,

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@ramarivera/coding-agent-langfuse",
-  "version": "0.1.39",
+  "version": "0.1.41",
   "description": "Universal coding-agent Langfuse backfiller and live OTLP helpers",
   "type": "module",
   "license": "MIT",