jev-gateway 0.1.0 → 0.2.0

package/dist/index.js

@@ -3,6 +3,7 @@ import { TypeSafeClient } from "@typesafe-ai/sdk";
 import { createApp } from "./app.js";
 import { loadConfig } from "./config.js";
 import { createDump } from "./debug.js";
+import { createEventLog } from "./events.js";
 const config = loadConfig();
 // Reads TYPESAFE_API_KEY. One fast retry only: past that, failing open to the LLM is quicker.
 const jev = new TypeSafeClient({
@@ -14,8 +15,10 @@ const app = createApp({
     config,
     askJev: (request) => jev.systemOne(request),
     dump: createDump(config.debugDumpDir),
+    events: createEventLog({ historyFile: config.logFile }),
     log: (entry) => console.log(JSON.stringify({ time: new Date().toISOString(), ...entry })),
 });
-serve({ fetch: app.fetch, port: config.port }, ({ port }) => {
+serve({ fetch: app.fetch, hostname: config.host, port: config.port }, ({ port }) => {
     console.log(`jev-gateway listening on http://localhost:${port} → ${config.upstreamBaseUrl} (jev: ${config.jevModel})`);
+    console.log(`dashboard: http://localhost:${port}/dashboard`);
 });

package/dist/usage.js

@@ -0,0 +1,86 @@
+const num = (value) => (typeof value === "number" && Number.isFinite(value) ? value : 0);
+const obj = (value) => (value && typeof value === "object" ? value : {});
+/** Fold one provider `usage` object into the running total; later reports override earlier ones. */
+function merge(into, raw) {
+    if ("prompt_tokens" in raw || "completion_tokens" in raw) {
+        // Chat Completions
+        into.input = num(raw.prompt_tokens);
+        into.output = num(raw.completion_tokens);
+        into.cached = num(obj(raw.prompt_tokens_details).cached_tokens);
+        into.reasoning = num(obj(raw.completion_tokens_details).reasoning_tokens);
+    }
+    else if ("cache_read_input_tokens" in raw || "cache_creation_input_tokens" in raw) {
+        // Anthropic, message_start: the input side, with a placeholder output count.
+        into.cached = num(raw.cache_read_input_tokens);
+        into.cacheWrite = num(raw.cache_creation_input_tokens);
+        into.input = num(raw.input_tokens) + into.cached + into.cacheWrite;
+        into.output = num(raw.output_tokens);
+    }
+    else {
+        // Responses API — or Anthropic's message_delta, which only updates the output count.
+        if ("input_tokens" in raw) {
+            into.input = num(raw.input_tokens);
+            into.cached = num(obj(raw.input_tokens_details).cached_tokens);
+        }
+        if ("output_tokens" in raw) {
+            into.output = num(raw.output_tokens);
+            into.reasoning = num(obj(raw.output_tokens_details).reasoning_tokens);
+        }
+    }
+}
+function collect(payload, into) {
+    const root = obj(payload);
+    // Where each API keeps it: top level (JSON replies, chat chunks, message_delta),
+    // `response.usage` (Responses events), `message.usage` (Anthropic message_start).
+    for (const holder of [root, obj(root.response), obj(root.message)]) {
+        if (holder.usage && typeof holder.usage === "object")
+            merge(into, holder.usage);
+    }
+}
+/**
+ * Read a reply to its end and report what it cost. Works on a clone, in the background: the
+ * client's own stream is never delayed. A stream cut short (Codex hangs up as soon as it has
+ * `response.completed`) still yields whatever usage arrived before the cut.
+ */
+export async function readUsage(response) {
+    const found = {};
+    const streaming = response.headers.get("content-type")?.includes("text/event-stream") ?? false;
+    let pending = "";
+    const scan = (line) => {
+        // Most stream events are text deltas; only the few that mention usage are worth parsing.
+        if (!line.startsWith("data:") || !line.includes('"usage"'))
+            return;
+        try {
+            collect(JSON.parse(line.slice(5)), found);
+        }
+        catch {
+            // A line cut short by the client hanging up.
+        }
+    };
+    try {
+        for await (const chunk of response.body?.pipeThrough(new TextDecoderStream()) ?? []) {
+            pending += chunk;
+            if (!streaming)
+                continue;
+            const lines = pending.split("\n");
+            pending = lines.pop() ?? "";
+            lines.forEach(scan);
+        }
+    }
+    catch {
+        // Aborted mid-stream: keep what was seen.
+    }
+    if (streaming)
+        scan(pending);
+    else {
+        try {
+            collect(JSON.parse(pending), found);
+        }
+        catch {
+            // Not JSON (an HTML error page, an empty body): nothing to report.
+        }
+    }
+    if (found.input === undefined && found.output === undefined)
+        return undefined;
+    return { input: 0, output: 0, cached: 0, cacheWrite: 0, reasoning: 0, ...found };
+}

package/package.json

@@ -1,7 +1,7 @@
 {
   "name": "jev-gateway",
-  "version": "0.1.0",
-  "description": "LLM gateway that hands tool-selection decisions to TypeSafe's Jev model \u2014 with launchers for Codex and Claude Code",
+  "version": "0.2.0",
+  "description": "Local LLM gateway that lets TypeSafe's Jev model pick the tool for coding agents. Launchers for Codex and Claude Code, with a monitoring dashboard.",
   "keywords": [
     "llm",
     "gateway",
@@ -16,9 +16,17 @@
   ],
   "license": "MIT",
   "author": "Vinicius Lana",
+  "repository": {
+    "type": "git",
+    "url": "git+https://github.com/vinilana/jev-gateway.git"
+  },
+  "homepage": "https://github.com/vinilana/jev-gateway#readme",
+  "bugs": {
+    "url": "https://github.com/vinilana/jev-gateway/issues"
+  },
   "type": "module",
   "engines": {
-    "node": ">=22.9"
+    "node": ">=22.15"
   },
   "bin": {
     "jev-codex": "bin/jev-codex.mjs",
@@ -28,12 +36,13 @@
     "dist",
     "bin",
     "scripts/mock-jev.mjs",
-    ".env.example"
+    ".env.example",
+    "CHANGELOG.md"
   ],
   "scripts": {
     "dev": "tsx watch --env-file-if-exists=.env src/index.ts",
     "start": "node --env-file-if-exists=.env dist/index.js",
-    "build": "rm -rf dist && tsc -p tsconfig.build.json",
+    "build": "rm -rf dist && tsc -p tsconfig.build.json && cp src/dashboard.html dist/",
     "typecheck": "tsc --noEmit",
     "test": "vitest run",
     "codex": "node bin/jev-codex.mjs",