@jaralabs/codex-bridge 0.4.3 → 0.4.5

package/README.md

@@ -1,5 +1,15 @@
 # @jaralabs/codex-bridge
 
+> **Disclaimer.** This project is not affiliated with, endorsed by, or
+> sponsored by OpenAI. It is an independent tool that reuses the OAuth
+> session written by the official Codex CLI to `~/.codex/auth.json` and
+> speaks the same upstream endpoint (`chatgpt.com/backend-api/codex/responses`)
+> that the CLI itself uses. The endpoint is not part of any published OpenAI
+> API, so the bridge can stop working the day OpenAI rotates it. Use it as
+> a developer convenience for your own account, not as a service you sell
+> or redistribute. The authors and contributors are not responsible for
+> account actions taken by OpenAI as a result of how you use this tool.
+
 > Use your ChatGPT Business / Team / Plus subscription from OpenCode instead
 > of paying for the OpenAI API.
 

package/lib/resilience.mjs

@@ -31,6 +31,15 @@ const state = {
   lastError: null,
   lastSuccessAt: 0,
   lastFailureAt: 0,
+  // token usage accumulated since process start. The bridge reads
+  // `usage` (input_tokens / output_tokens / total_tokens) out of every
+  // successful Responses API response and sums it here. This is the
+  // closest thing the bridge has to a "credits used" counter: the
+  // ChatGPT subscription does not expose a per-account quota, so
+  // the only observable usage is what flows through the proxy.
+  promptTokens: 0,
+  completionTokens: 0,
+  totalTokens: 0,
   // circuit breaker
   failureTimes: [],   // timestamps of recent failures
   circuitOpenedAt: 0, // 0 = closed; >0 = open since this timestamp
@@ -116,11 +125,33 @@ export function metrics() {
     lastError: state.lastError,
     lastSuccessAt: state.lastSuccessAt,
     lastFailureAt: state.lastFailureAt,
+    tokens: {
+      prompt: state.promptTokens,
+      completion: state.completionTokens,
+      total: state.totalTokens,
+    },
     circuit: circuitStatus(),
     circuitFailuresInWindow: state.failureTimes.length,
   };
 }
 
+// Sum the token counts from a successful Responses API response
+// into the in-memory counters. The wire shape is
+// `{ input_tokens, output_tokens, total_tokens }`; missing fields
+// default to 0. Idempotent and safe to call multiple times for the
+// same response (callers should only call once per success).
+export function recordUsage(usage) {
+  if (!usage || typeof usage !== "object") return;
+  const prompt = Number.isFinite(usage.input_tokens) ? usage.input_tokens : 0;
+  const completion = Number.isFinite(usage.output_tokens) ? usage.output_tokens : 0;
+  const total = Number.isFinite(usage.total_tokens)
+    ? usage.total_tokens
+    : prompt + completion;
+  state.promptTokens += prompt;
+  state.completionTokens += completion;
+  state.totalTokens += total;
+}
+
 // Wrap an async function with retry + circuit breaker semantics.
 // fn() must throw an Error with .status attached when the response
 // is a retryable HTTP error. onRetry(attempt, delayMs, err) is

package/lib/server.mjs

@@ -30,6 +30,7 @@ import {
 import {
   withResilience,
   metrics as resilienceMetrics,
+  recordUsage,
 } from "./resilience.mjs";
 import { probeCapabilities } from "./capabilities.mjs";
 
@@ -440,6 +441,13 @@ async function handleChatCompletions(payload, req, res) {
   const id = genId("chatcmpl");
   const created = now();
   const usage = toUsage(result.usage);
+  // Sum the token usage of this successful turn into the in-memory
+  // counters that /v1/health exposes. We record once per turn (here,
+  // after the upstream Responses API returned a `response.completed`
+  // event with a `usage` block), so streaming and JSON responses are
+  // counted identically. Multi-turn tool-call loops in the same
+  // request accumulate multiple `recordUsage` calls.
+  if (result.usage) recordUsage(result.usage);
   const hasToolCalls = result.toolCalls && result.toolCalls.length > 0;
   const finishReason = hasToolCalls ? "tool_calls" : "stop";
   const message = { role: "assistant", content: result.text || "" };

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@jaralabs/codex-bridge",
-  "version": "0.4.3",
+  "version": "0.4.5",
   "private": false,
   "description": "OpenAI-compatible HTTP bridge that reuses a local ChatGPT Business / Team / Plus session via codex, so OpenCode can list and call those models without paying for the OpenAI API.",
   "type": "module",