@descix/cli 2.0.4 → 3.0.0

package/bin/descix.js

@@ -7,7 +7,7 @@
  * Combines CLI and MCP functionality with unified authentication.
  */
 
-import { Command } from 'commander';
+import { Command, Option } from 'commander';
 import chalk from 'chalk';
 import { DeSciXApiClient } from '../lib/api-client.js';
 import { requireAuth } from '../lib/auth-guard.js';
@@ -1663,7 +1663,8 @@ appCommand
 // ============ App: set-default-model (WS-CONFIG-BOOTSTRAP-FIX item #10) ============
 //
 // Updates Community/{c}/Apps/{a}.default_app_model. Per CEO inheritance chain:
-//   options.model > kb.kb_model_override > app.default_app_model > levelConfig.model > DEFAULT_AI_MODEL
+//   request intelligence_level > kb.kb_model_override > app.default_app_model > the level's model
+// (there is no separate default model: the default is the model of DEFAULT_INTELLIGENCE_LEVEL)
 //
 // --clear deletes the field via FieldValue.delete() (NOT null) so the resolver falls through
 // cleanly to per-level platform defaults.
@@ -1672,13 +1673,13 @@ appCommand
   .command('set-default-model')
   .description('Set or clear App.default_app_model (model used when no KB override is set)')
   .requiredOption('-a, --app <app_id>', 'App ID (e.g. unk-cos)')
-  .option('-m, --model <model_name>', 'Gemini model name (e.g. gemini-3.1-flash-lite). Mutually exclusive with --clear.')
+  .option('-m, --model <model_name>', 'Model name (e.g. gemini-3.8-flash). Mutually exclusive with --clear.')
   .option('--clear', 'Delete App.default_app_model via FieldValue.delete() (resets to platform-default inheritance)')
   .addHelpText('after', `
-NOTE: Pinning a "Pro" model (e.g., gemini-2.5-pro, gemini-3.1-pro-preview) as
-an app or KB override will fail at L1 — these models require thinking mode and
-reject L1's thinkingBudget=0. If you intend an L1-compatible override, use a
-"Flash" model (gemini-2.5-flash, gemini-3.1-flash-lite, gemini-3-flash-preview).
+NOTE: A pinned model is paired with a thinking_level, and which levels exist is a
+property of the MODEL. The server refuses, by name, a (model, thinking_level) pair
+it does not support. The current intelligence levels (each names its own model and
+thinking_level) are printed under "Per-level" after this command runs.
 `)
   .action(async (options) => {
     try {
@@ -1820,12 +1821,12 @@ kbCommand
   .description('Set KB.kb_model_override (per-KB model selection)')
   .requiredOption('-a, --app <app_id>', 'App ID')
   .requiredOption('-k, --kb <kb_name>', 'KB name')
-  .requiredOption('-m, --model <model_name>', 'Gemini model name (e.g. gemini-2.5-pro)')
+  .requiredOption('-m, --model <model_name>', 'Model name (e.g. gemini-3.8-flash)')
   .addHelpText('after', `
-NOTE: Pinning a "Pro" model (e.g., gemini-2.5-pro, gemini-3.1-pro-preview) as
-an app or KB override will fail at L1 — these models require thinking mode and
-reject L1's thinkingBudget=0. If you intend an L1-compatible override, use a
-"Flash" model (gemini-2.5-flash, gemini-3.1-flash-lite, gemini-3-flash-preview).
+NOTE: A pinned model is paired with a thinking_level, and which levels exist is a
+property of the MODEL. The server refuses, by name, a (model, thinking_level) pair
+it does not support. The current intelligence levels (each names its own model and
+thinking_level) are printed under "Per-level" after this command runs.
 `)
   .action(async (options) => {
     try {
@@ -3360,11 +3361,15 @@ program
   .option('--apps <ids>', 'Comma-separated app IDs for cross-app query')
   .option('--level <n>', 'Intelligence level (1-5)', parseInt)
   .option('--model <name>', 'Explicit model override')
-  .option('--thinking <budget>', 'Thinking token budget (-1=dynamic, 0=off, N=fixed)', parseInt)
+  .option('--thinking-level <level>', 'Thinking level for this call: minimal, low, medium or high. The server refuses a level the resolved model does not support.')
+  .addOption(new Option('--thinking <budget>').hideHelp())
   .option('--tokens', 'Show token usage in response footer')
   .option('--new', 'Start a new conversation (clear session)')
   .action(async (questionArgs, options) => {
     try {
+      if (options.thinking !== undefined) {
+        throw new Error('--thinking (a token budget) was removed. Use --thinking-level <level>, one of minimal, low, medium, high.');
+      }
       const apiClient = new DeSciXApiClient();
       await requireAuth(apiClient);
 
@@ -3390,7 +3395,7 @@ program
           user_input: question,
           intelligence_level: options.level,
           model: options.model,
-          thinking_budget: options.thinking,
+          thinking_level: options.thinkingLevel,
           streaming: false
         });
         const result = response.message || response;
@@ -3497,7 +3502,7 @@ program
         streaming: false,
         intelligence_level: options.level,
         model: options.model,
-        thinking_budget: options.thinking,
+        thinking_level: options.thinkingLevel,
       };
 
       if (useMultiKb) {

package/lib/commands/model-config.js

@@ -7,8 +7,9 @@
  *   - descix kb set-override-model -a <app_id> -k <kb_name> -m <model_name> [--env=dev|demo|prod]
  *   - descix kb clear-override-model -a <app_id> -k <kb_name> [--env=dev|demo|prod]
  *
- * Per CEO 2026-05-26 inheritance chain:
- *   options.model || kb.kb_model_override || app.default_app_model || levelConfig.model || utils.DEFAULT_AI_MODEL
+ * Resolver order (the server owns it; this CLI prints the chain the server reports):
+ *   request intelligence_level > kb.kb_model_override (+ its thinking_level) > app.default_app_model > the level's model
+ * There is no separate default model: the default is the model of the platform's default intelligence level.
  *
  * --clear flows use FieldValue.delete() server-side (NOT null) per tripwire #2.
  *
@@ -110,12 +111,10 @@ function identityFromApiClient(apiClient) {
 function printResolution({ app_id, resolution }) {
     if (!resolution) return;
     const dam = resolution.default_app_model === null ? 'null' : `"${resolution.default_app_model}"`;
-    const dai = resolution.default_ai_model ? `"${resolution.default_ai_model}"` : 'null';
 
     console.log(chalk.cyan('\nResolution chain that will apply going forward:'));
     console.log(chalk.gray(`  ${resolution.chain}`));
     console.log(chalk.gray(`  app.default_app_model = ${dam}`));
-    console.log(chalk.gray(`  utils.DEFAULT_AI_MODEL = ${dai}`));
 
     const levels = resolution.intelligence_levels || {};
     const levelKeys = Object.keys(levels).sort();

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@descix/cli",
-  "version": "2.0.4",
+  "version": "3.0.0",
   "description": "DeSciX CLI and MCP Server - Unified command-line interface for DeSciX platform",
   "repository": {
     "type": "git",
@@ -40,7 +40,7 @@
     "googleapis": "^170.1.0",
     "inquirer": "^12.1.0",
     "ora": "^9.0.0",
-    "@descix/platform-api": "^2.0.0"
+    "@descix/platform-api": "^4.0.0"
   },
   "devDependencies": {
     "acorn": "^8.18.0"

package/templates/agent-claude.md

@@ -288,7 +288,7 @@ You are bonded to cloud knowledge bases. **BEFORE starting any task**, query the
 
 **Multi-app query (preferred — richer context):**
 ```bash
-descix chat --apps APP1,APP2 -q "YOUR SPECIFIC QUESTION" --level 3 --thinking -1 --tokens
+descix chat --apps APP1,APP2 -q "YOUR SPECIFIC QUESTION" --level 3 --thinking-level low --tokens
 ```
 
 **Single-app query:**
@@ -303,7 +303,7 @@ ask_question_to_app({ app_id: "APP_ID", user_input: "YOUR SPECIFIC QUESTION" })
 
 **Query parameters:**
 - `--level 1-5` — intelligence (1=fast lookup, 3=balanced, 5=deep reasoning)
-- `--thinking -1` — dynamic thinking budget; `0`=off, `N`=fixed tokens
+- `--thinking-level minimal|low|medium|high` — how much the model thinks; the server refuses a level the resolved model does not support
 - `--tokens` — show token usage in response
 - Omit `--new` to continue the session thread (Gemini remembers context)