@agentprojectcontext/apx 1.65.2 → 1.66.0

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@agentprojectcontext/apx",
-  "version": "1.65.2",
+  "version": "1.66.0",
   "description": "APX — unified CLI + daemon for the Agent Project Context (APC) standard.",
   "publishConfig": {
     "access": "public"

package/src/core/config/index.js

@@ -152,7 +152,7 @@ const DEFAULT_CONFIG = {
       piper:      { bin: "piper", model: "", speaker: "", extra_args: [] },
       elevenlabs: { api_key: "", model: "eleven_multilingual_v2", voice_id: "", output_format: "mp3_44100_128" },
       openai:     { api_key: "", model: "tts-1", voice: "alloy", format: "mp3" },
-      gemini:     { api_key: "", model: "gemini-2.5-flash-preview-tts", voice: "Kore" },
+      gemini:     { api_key: "", model: "gemini-2.5-flash-tts", voice: "Kore" },
     },
   },
 };

package/src/core/engines/presets.js

@@ -0,0 +1,102 @@
+// Curated engine catalog — the SINGLE SOURCE OF TRUTH for the known models,
+// per-engine defaults, base URLs and api-key env vars shown across APX.
+//
+// Who consumes this:
+//   • CLI  — `apx setup` builds its provider/model menus from here
+//     (src/interfaces/cli/commands/setup.js).
+//   • Web  — the admin panel fetches it via `GET /engines/presets` and hydrates
+//     its provider forms (src/interfaces/web/.../providers/typeStyles.ts).
+//
+// This is the OFFLINE / no-key fallback list. When the user has an api_key
+// configured, the daemon's `POST /engines/models` returns the provider's LIVE
+// catalog instead (see ./catalog.js). The model field is ALWAYS free-text, so
+// any id can be typed even if it is not listed here.
+//
+// `ollama` and `custom` are intentionally dynamic — no curated model list.
+// Update model ids in THIS file only; every surface reflects the change.
+
+/** @typedef {{ base_url: string, default_model: string, api_key_env: string, known_models: string[] }} EnginePreset */
+
+/** @type {Record<string, EnginePreset>} */
+export const ENGINE_PRESETS = {
+  anthropic: {
+    base_url: "", // empty ⇒ adapter uses the built-in Anthropic endpoint
+    default_model: "claude-sonnet-5",
+    api_key_env: "ANTHROPIC_API_KEY",
+    known_models: [
+      "claude-opus-4-8",
+      "claude-sonnet-5",
+      "claude-haiku-4-5",
+      "claude-fable-5",
+    ],
+  },
+  openai: {
+    base_url: "https://api.openai.com/v1",
+    default_model: "gpt-5.4-mini",
+    api_key_env: "OPENAI_API_KEY",
+    known_models: [
+      "gpt-5.5",
+      "gpt-5.4-mini",
+      "gpt-5.4-nano",
+      "gpt-5.1",
+      "gpt-4.1-mini",
+    ],
+  },
+  gemini: {
+    base_url: "https://generativelanguage.googleapis.com/v1beta/openai",
+    default_model: "gemini-2.5-flash",
+    api_key_env: "GEMINI_API_KEY",
+    known_models: [
+      "gemini-3.5-flash",
+      "gemini-3.1-pro-preview",
+      "gemini-2.5-pro",
+      "gemini-2.5-flash",
+      "gemini-2.5-flash-lite",
+    ],
+  },
+  groq: {
+    base_url: "https://api.groq.com/openai/v1",
+    default_model: "openai/gpt-oss-20b",
+    api_key_env: "GROQ_API_KEY",
+    known_models: [
+      "openai/gpt-oss-120b",
+      "openai/gpt-oss-20b",
+      "qwen/qwen3.6-27b",
+      "groq/compound",
+      "groq/compound-mini",
+      "whisper-large-v3-turbo",
+    ],
+  },
+  openrouter: {
+    base_url: "https://openrouter.ai/api/v1",
+    // openrouter/auto = "Auto Router": OpenRouter picks the best model.
+    default_model: "openrouter/auto",
+    api_key_env: "OPENROUTER_API_KEY",
+    known_models: [
+      "openrouter/auto",
+      "openrouter/free",
+      "anthropic/claude-sonnet-5",
+      "openai/gpt-5.4-mini",
+      "google/gemini-2.5-flash",
+    ],
+  },
+  azure: {
+    base_url: "",
+    default_model: "", // Azure deployment names are user-defined
+    api_key_env: "AZURE_OPENAI_API_KEY",
+    known_models: [],
+  },
+  ollama: {
+    base_url: "http://127.0.0.1:11434",
+    default_model: "gemma2:9b",
+    api_key_env: "",
+    known_models: [], // dynamic — fetched live from the local Ollama daemon
+  },
+  mock: { base_url: "", default_model: "mock", api_key_env: "", known_models: ["mock"] },
+  custom: { base_url: "", default_model: "", api_key_env: "", known_models: [] },
+};
+
+/** Known models for one engine, or [] if the engine is dynamic/unknown. */
+export function knownModels(engine) {
+  return ENGINE_PRESETS[engine]?.known_models ?? [];
+}

package/src/core/voice/engines/gemini.js

@@ -1,13 +1,12 @@
 // Gemini TTS adapter.
-// At the time of writing the Gemini TTS surface is gated and not stable across
-// SDK versions: some models (gemini-2.5-flash-preview-tts) expose synthesize
-// via the v1beta REST surface, others require Vertex. To keep APX engine-
-// agnostic, this adapter performs a best-effort call against the documented
-// REST shape, but flags itself as not-implemented when the response does not
-// include inline audio data.
+// The Gemini TTS surface can vary across SDK versions: some models
+// (gemini-2.5-flash-tts) expose synthesize via the v1beta REST surface, others
+// require Vertex. To keep APX engine-agnostic, this adapter performs a
+// best-effort call against the documented REST shape, but flags itself as
+// not-implemented when the response does not include inline audio data.
 //
 // Config (~/.apx/config.json → voice.tts.gemini):
-//   { "api_key": "...", "model": "gemini-2.5-flash-preview-tts", "voice": "Kore",
+//   { "api_key": "...", "model": "gemini-2.5-flash-tts", "voice": "Kore",
 //     "style": "habla en tono alegre y enérgico" }
 //
 // `style` is an optional natural-language instruction describing HOW the voice
@@ -22,7 +21,7 @@ import fs from "node:fs";
 import path from "node:path";
 import { randomUUID } from "node:crypto";
 
-const DEFAULT_MODEL = "gemini-2.5-flash-preview-tts";
+const DEFAULT_MODEL = "gemini-2.5-flash-tts";
 
 function getKey(config, parentEnginesCfg) {
   return (

package/src/host/daemon/api/engines.js

@@ -1,12 +1,18 @@
 // GET  /engines            — list engine adapter ids known to core/engines.
+// GET  /engines/presets     — curated catalog (known models, defaults) per engine.
 // POST /engines/models      — live model catalog from a provider.
 // GET  /engines/models      — legacy (Ollama only, no auth).
 import { ENGINE_IDS } from "#core/engines/index.js";
 import { listModels } from "#core/engines/catalog.js";
+import { ENGINE_PRESETS } from "#core/engines/presets.js";
 
 export function register(app, { config }) {
   app.get("/engines", (_req, res) => res.json({ engines: ENGINE_IDS }));
 
+  // Curated fallback catalog shared with the CLI wizard. The web hydrates its
+  // provider forms from here so model lists never drift between surfaces.
+  app.get("/engines/presets", (_req, res) => res.json({ presets: ENGINE_PRESETS }));
+
   app.post("/engines/models", async (req, res) => {
     const b = req.body || {};
     const engine = String(b.engine || "").toLowerCase();

package/src/interfaces/cli/commands/setup.js

@@ -8,6 +8,7 @@ import http from "node:http";
 import readline from "node:readline";
 import { spawnSync } from "node:child_process";
 import { readConfig, writeConfig } from "#core/config/index.js";
+import { ENGINE_PRESETS } from "#core/engines/presets.js";
 import { mascot } from "#core/mascot.js";
 import { setupClaudePermissions } from "../claude-permissions.js";
 import { PERMISSION_MODES, DEFAULT_PERMISSION_MODE } from "#core/constants/permissions.js";
@@ -59,6 +60,8 @@ async function fetchOllamaModels(baseUrl) {
 }
 
 // ── Provider definitions ──────────────────────────────────────────────────────
+// Model lists come from the shared catalog (#core/engines/presets.js) so the CLI
+// and the web admin panel never drift. Ollama stays dynamic (fetched at runtime).
 const PROVIDERS = [
   {
     id: "anthropic",
@@ -66,7 +69,7 @@ const PROVIDERS = [
     needsKey: true,
     keyLabel: "Anthropic API key",
     keyHint: "sk-ant-...",
-    models: ["claude-sonnet-4-5", "claude-haiku-4-5", "claude-opus-4-5"],
+    models: ENGINE_PRESETS.anthropic.known_models,
   },
   {
     id: "openai",
@@ -74,7 +77,7 @@ const PROVIDERS = [
     needsKey: true,
     keyLabel: "OpenAI API key",
     keyHint: "sk-...",
-    models: ["gpt-4o", "gpt-4o-mini", "gpt-4-turbo"],
+    models: ENGINE_PRESETS.openai.known_models,
   },
   {
     id: "ollama",
@@ -88,7 +91,7 @@ const PROVIDERS = [
     needsKey: true,
     keyLabel: "Gemini API key",
     keyHint: "AIza...",
-    models: ["gemini-2.0-flash", "gemini-1.5-pro"],
+    models: ENGINE_PRESETS.gemini.known_models,
   },
 ];
 
@@ -254,7 +257,7 @@ export async function cmdSetup() {
     console.log(`    ${cy("2")}. piper      ${di("(local, offline; needs piper CLI + voice model)")}`);
     console.log(`    ${cy("3")}. elevenlabs ${di("(cloud; eleven_multilingual_v2)")}`);
     console.log(`    ${cy("4")}. openai     ${di("(cloud; tts-1, reuses your openai key)")}`);
-    console.log(`    ${cy("5")}. gemini     ${di("(experimental, best-effort)")}`);
+    console.log(`    ${cy("5")}. gemini     ${di("(cloud; gemini-2.5-flash-tts, 30 voices)")}`);
     console.log(`    ${cy("6")}. mock       ${di("(silent WAV; useful for tests)")}`);
     console.log();
     const choice = (await ask(`  Choose [1-6, default 1]: `)).trim() || "1";