bizrouter 0.1.0 → 0.1.1

package/dist/commands/misc.js

@@ -2,7 +2,7 @@ import { spawn } from 'node:child_process';
 import { ENV_KEY_NAME } from '../config.js';
 import { c, info, print } from '../ui.js';
 import { requireApiKey } from './shared.js';
-export const VERSION = '0.1.0';
+export const VERSION = '0.1.1';
 /** `eval "$(bizrouter env)"` exports the saved key for tools configured by `bizrouter setup`. */
 export function envCommand() {
     const key = requireApiKey();

package/dist/harness/hermes.js

@@ -1,3 +1,15 @@
+/** Hermes sends this max_tokens to custom OpenAI-compatible endpoints regardless of the model. */
+export const HERMES_FIXED_MAX_TOKENS = 65536;
+/**
+ * Models whose output cap is below Hermes' fixed max_tokens get rejected by the
+ * gateway (`max_tokens: 65536 > 64000 …`) and Hermes misreads that as a context
+ * overflow, so warn before launching.
+ */
+export function hermesOutputCapWarning(model, maxOutputTokens) {
+    if (maxOutputTokens === undefined || maxOutputTokens >= HERMES_FIXED_MAX_TOKENS)
+        return undefined;
+    return `Hermes 는 max_tokens ${HERMES_FIXED_MAX_TOKENS.toLocaleString('en-US')} 을 고정으로 보내는데 ${model} 의 출력 한도는 ${maxOutputTokens.toLocaleString('en-US')} 입니다. 요청이 400 으로 거절되고 Hermes 가 이를 컨텍스트 초과로 오해합니다. 출력 한도가 65,536 이상인 모델(예: anthropic/claude-sonnet-5)을 쓰세요.`;
+}
 const HERMES_SUBCOMMANDS = new Set(['chat']);
 /**
  * Hermes reaches any OpenAI-compatible endpoint through its `custom` provider,

package/dist/index.js

@@ -13,7 +13,7 @@ import { buildClaudePlan } from './harness/claude.js';
 import { buildCodexPlan, codexConfigPinsServiceTier } from './harness/codex.js';
 import { codexConfigPath } from './commands/setup.js';
 import { existsSync, readFileSync } from 'node:fs';
-import { buildHermesPlan } from './harness/hermes.js';
+import { buildHermesPlan, hermesOutputCapWarning } from './harness/hermes.js';
 import { buildOpencodePlan } from './harness/opencode.js';
 import { runPlan } from './spawn.js';
 import { CliError, fail, info, maskKey, print, warn } from './ui.js';
@@ -93,6 +93,12 @@ async function launch(harness, argv) {
             plan = buildHermesPlan({ apiKey, apiBase: base, model: model ?? FALLBACK_DEFAULT.hermes, passthrough: args.passthrough });
             if (args.reasoningEffort)
                 warn('Hermes 는 추론 강도 옵션을 받지 않습니다.');
+            {
+                const hermesModel = model ?? FALLBACK_DEFAULT.hermes;
+                const capWarning = hermesOutputCapWarning(hermesModel, models.find((m) => m.id === hermesModel)?.max_output_tokens);
+                if (capWarning)
+                    warn(capWarning);
+            }
             break;
     }
     if (!args.dryRun)

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "bizrouter",
-  "version": "0.1.0",
+  "version": "0.1.1",
   "description": "BizRouter CLI - run Claude Code, Codex, OpenCode, and Hermes through BizRouter with one command",
   "license": "UNLICENSED",
   "type": "module",