ag-ai 0.1.6 → 0.2.0

package/README.md

@@ -9,13 +9,17 @@ Provider-neutral AI client and cluster runtime for the
 | --------------- | --------------------------------------------------------------------- |
 | `ag-ai`         | Client SDK: generate, generateText, promptDirect, promptImage, models |
 | `ag-ai/client`  | Client SDK entry point                                                |
-| `ag-ai/runtime` | Cluster runtime: generation protocol, store, server, bridge (ESM)     |
 | `ag-ai/k8s`     | Cluster plugin descriptor (`ClusterPlugin`)                           |
+| `ag-ai/runtime` | Cluster runtime: generation protocol, store, server, bridge (ESM)     |
 | `ag-ai/image`   | Neutral image-output adapter (provider per implementation)            |
 
 The client compiles to CommonJS. The runtime is hand-authored ESM and shipped
 as-is, because it runs inside the pod with no build step.
 
+The client-only dependencies (`@google/genai`, `node-cache`) are declared as
+`optionalDependencies`; the cluster runtime install omits them because the
+runtime imports only Node builtins.
+
 ## Boundaries
 
 - No prompts. Prompt templates belong to consumers (gec.dev, bolt.garden, …).

package/dist/client/adapters/google.js

@@ -390,8 +390,7 @@ const generateGoogle = async (request, config = {}) => {
                 // Reactive per-model backoff: a 429 skips the model until its window
                 // elapses while every other model keeps serving.
                 if ((0, quota_1.isModelBackedOff)(selectedModel)) {
-                    (0, log_1.debug)('skipping rate-limited model', {
-                        model: selectedModel,
+                    (0, log_1.debug)(`[${selectedModel}] skipping rate-limited model`, {
                         remainingMs: (0, quota_1.modelBackoffRemainingMs)(selectedModel),
                     });
                     continue;
@@ -400,7 +399,7 @@ const generateGoogle = async (request, config = {}) => {
                     ...requestConfig,
                     model: selectedModel,
                 };
-                (0, log_1.info)('generation request', { model: selectedModel, ident: request.ident });
+                (0, log_1.info)(`[${selectedModel}] generation request`, { ident: request.ident });
                 try {
                     // oxlint-disable-next-line no-await-in-loop -- model fallback is sequential by design
                     const response = await awaitWithSignal(ai.models.generateContent(selectedRequest), controller.signal);
@@ -411,7 +410,7 @@ const generateGoogle = async (request, config = {}) => {
                         generatedAt: config.dependencies?.now?.() ?? Date.now(),
                         ...(usage === undefined ? {} : { usage }),
                     };
-                    (0, log_1.debug)('generation completed', request.ident);
+                    (0, log_1.debug)(`[${result.model}] generation completed`, request.ident);
                     return result;
                 }
                 catch (error) {
@@ -424,8 +423,7 @@ const generateGoogle = async (request, config = {}) => {
                     const unavailableModel = failureStatus === 404 && message.includes(selectedModel);
                     // Fallback only; per-model backoff below is keyed by rate limiting.
                     if (rateLimited || overloadedModel || unavailableModel) {
-                        (0, log_1.warn)('generation attempt failed; trying next available combination', {
-                            model: selectedModel,
+                        (0, log_1.warn)(`[${selectedModel}] generation attempt failed; trying next available combination`, {
                             status: failureStatus,
                         });
                         (0, apikey_1.blockKeyService)(key, `gemini-${selectedModel}`);

package/dist/index.d.ts

@@ -1,2 +1,3 @@
 export * from './client';
 export * from './internal/apikey';
+export * from './internal/log';

package/dist/index.js

@@ -16,3 +16,4 @@ var __exportStar = (this && this.__exportStar) || function(m, exports) {
 Object.defineProperty(exports, "__esModule", { value: true });
 __exportStar(require("./client"), exports);
 __exportStar(require("./internal/apikey"), exports);
+__exportStar(require("./internal/log"), exports);

package/dist/internal/apikey.js

@@ -33,7 +33,7 @@ const getAvailableCombinations = (service, resetIfExhausted = true) => {
     }
     // If all combinations are blocklisted, clear the blocklist and try again
     if (combinations.length === 0 && resetIfExhausted) {
-        (0, log_1.warn)(`All API key + service combinations were blocklisted for ${service}. Clearing blocklist and trying again.`);
+        (0, log_1.warn)(`[${service}] all API key + service combinations were blocklisted; clearing blocklist and trying again.`);
         blocklist.flushAll();
         for (const key of keys) {
             combinations.push({ service, key });
@@ -45,7 +45,7 @@ const getAvailableCombinations = (service, resetIfExhausted = true) => {
 exports.getAvailableCombinations = getAvailableCombinations;
 // Block a key+service combination
 const blockKeyService = (apiKey, service) => {
-    (0, log_1.warn)(`key+service blocklisted for ${service}`);
+    (0, log_1.warn)(`[${service}] key+service blocklisted`);
     addToBlocklist(apiKey, service);
 };
 exports.blockKeyService = blockKeyService;

package/dist/k8s/index.js

@@ -14,7 +14,7 @@ const t3code_1 = require("./t3code");
 exports.plugin = {
     name: 'ag-ai',
     namespace: common_1.NAMESPACE,
-    requiredEnv: ['AG_AI_QUEUE_URL', 'AG_AI_GENERATION_CIDR'],
+    requiredEnv: ['AG_AI_QUEUE_URL', 'AG_AI_GENERATION_CIDR', 'AG_AI_PACKAGE'],
     requiredSecrets: [
         'runner-token',
         'generation-token',

package/dist/k8s/runner.js

@@ -10,7 +10,6 @@ const ag_rpi_mega_1 = require("ag-rpi-mega");
 const common_1 = require("./common");
 const networkInitTemplate = (0, node_fs_1.readFileSync)(node_path_1.default.join(__dirname, 'assets', 'network-init.sh'), 'utf8');
 /** The published runtime is installed into a writable emptyDir at pod start. */
-const AG_AI_PACKAGE = 'ag-ai@^0.1.1';
 const AG_AI_RUNTIME_DIR = '/opt/ag-ai';
 const AG_AI_BRIDGE = `${AG_AI_RUNTIME_DIR}/node_modules/ag-ai/runtime/bridge.mjs`;
 const networkInitScript = ({ generationCidr, generationPort, opencodePort, }) => networkInitTemplate
@@ -41,6 +40,9 @@ const runnerStack = (context) => {
         throw new Error('AG_AI_GENERATION_CIDR is required');
     const imageProvider = context.env.AG_AI_IMAGE_PROVIDER ?? 'openai';
     const opencodeProvider = context.env.AG_AI_OPENCODE_PROVIDER;
+    const agAiPackage = context.env.AG_AI_PACKAGE;
+    if (!agAiPackage)
+        throw new Error('AG_AI_PACKAGE is required');
     const metalLbAddress = facilities.metalLb(common_1.METAL_LB_KEY);
     if (!metalLbAddress)
         throw new Error(`no MetalLB address is configured for ${common_1.METAL_LB_KEY}`);
@@ -96,8 +98,6 @@ const runnerStack = (context) => {
             (0, common_1.env)('AG_AI_RUNNER_ID', common_1.RUNNER_APP),
             (0, common_1.env)('AG_AI_WORKSPACE', '/data/workspace'),
             (0, common_1.env)('AG_AI_MAIN_WORKSPACE', '/data/workspace'),
-            (0, common_1.env)('AG_AI_GENERATED_IMAGES', '/data/generated-images'),
-            (0, common_1.env)('AG_AI_MAIN_GENERATED_IMAGES', '/data/generated-images'),
             { name: 'AG_AI_GENERATION_HOST', valueFrom: { fieldRef: { fieldPath: 'status.podIP' } } },
             (0, common_1.env)('AG_AI_GENERATION_PORT', String(common_1.GENERATION_PORT)),
             (0, common_1.env)('AG_AI_GENERATION_STORE', '/data/generation-jobs'),
@@ -168,6 +168,8 @@ const runnerStack = (context) => {
             name: 'install-ag-ai',
             image: facilities.image('ag-ai-runner'),
             command: ['npm'],
+            // The runtime imports only Node builtins. Client-only dependencies are
+            // optional, so omitting them keeps the pod install lean.
             args: [
                 'install',
                 '--prefix',
@@ -175,8 +177,9 @@ const runnerStack = (context) => {
                 '--no-audit',
                 '--no-fund',
                 '--omit=dev',
+                '--omit=optional',
                 '--ignore-scripts',
-                AG_AI_PACKAGE,
+                agAiPackage,
             ],
             env: [(0, common_1.env)('HOME', '/tmp'), (0, common_1.env)('NPM_CONFIG_CACHE', '/tmp/npm-cache')],
             volumeMounts: [

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "ag-ai",
-  "version": "0.1.6",
+  "version": "0.2.0",
   "description": "Provider-neutral AI client and cluster runtime for the rpi-mega cluster.",
   "license": "ISC",
   "author": "admin@gec.dev",
@@ -11,20 +11,20 @@
   "files": [
     "dist/**/*",
     "runtime/**/*",
+    "!runtime/tests",
     "README.md"
   ],
   "main": "./dist/index.js",
   "types": "./dist/index.d.ts",
   "exports": {
     ".": "./dist/index.js",
-    "./client": "./dist/index.js",
+    "./client": "./dist/client/index.js",
     "./k8s": "./dist/k8s/index.js",
-    "./runtime": "./runtime/index.mjs"
-  },
-  "dependencies": {
-    "@google/genai": "^2.19.0",
-    "node-cache": "^5.1.2"
+    "./runtime": "./runtime/index.mjs",
+    "./image": "./runtime/image/adapter.mjs",
+    "./package.json": "./package.json"
   },
+  "dependencies": {},
   "devDependencies": {
     "@kubernetes/client-node": "2.0.0",
     "@types/node": "^26.4.1",
@@ -42,6 +42,10 @@
       "optional": true
     }
   },
+  "optionalDependencies": {
+    "@google/genai": "^2.19.0",
+    "node-cache": "^5.1.2"
+  },
   "scripts": {
     "preinstall": "npx only-allow pnpm",
     "format": "e7-oxfmt",

package/runtime/bridge.mjs

@@ -19,8 +19,6 @@ const DEFAULT_QUEUE_URL = 'https://queue.invalid/api/ai/runner';
 const DEFAULT_TOKEN_FILE = '/var/run/secrets/ag-ai-runner/token';
 const DEFAULT_WORKSPACE = '/workspace';
 const DEFAULT_MAIN_WORKSPACE = '/data/workspace';
-const DEFAULT_GENERATED_IMAGES = '/generated-images';
-const DEFAULT_MAIN_GENERATED_IMAGES = '/data/ag-ai/generated_images';
 const DEFAULT_GENERATION_PORT = 4501;
 const DEFAULT_GENERATION_STORE = '/generation-jobs';
 const DEFAULT_GENERATION_TOKEN_FILE = '/var/run/secrets/ag-ai-generation/token';
@@ -824,10 +822,6 @@ const readConfig = async (env = process.env) => {
     runnerId: env.AG_AI_RUNNER_ID ?? `ag-ai-${os.hostname()}-${randomUUID()}`,
     workspace: path.resolve(env.AG_AI_WORKSPACE ?? DEFAULT_WORKSPACE),
     mainWorkspace: path.resolve(env.AG_AI_MAIN_WORKSPACE ?? DEFAULT_MAIN_WORKSPACE),
-    generatedImages: path.resolve(env.AG_AI_GENERATED_IMAGES ?? DEFAULT_GENERATED_IMAGES),
-    mainGeneratedImages: path.resolve(
-      env.AG_AI_MAIN_GENERATED_IMAGES ?? DEFAULT_MAIN_GENERATED_IMAGES,
-    ),
     generationHost: env.AG_AI_GENERATION_HOST,
     generationPeerPrefix: env.AG_AI_GENERATION_PEER_PREFIX,
     generationPort: env.AG_AI_GENERATION_PORT ?? DEFAULT_GENERATION_PORT,
@@ -847,7 +841,6 @@ const main = async () => {
   const analytics = new AnalyticaEvents({ token: config.analyticaToken, source: config.runnerId });
   void analytics.emit('pod', 'started');
   await safeRoot(config.workspace, 'workspace root');
-  await safeRoot(config.generatedImages, 'generated-image root');
 
   const imageApiKey = await readOptionalToken(config.imageApiKeyFile);
   let imageAdapter;

package/runtime/index.mjs

@@ -9,7 +9,6 @@ export {
   normalizeModelCatalog,
   publicModelCatalog,
   requestDigest,
-  toRunnerInput,
 } from './protocol.mjs';
 export { GenerationStore, GenerationStoreError } from './store.mjs';
 export {

package/runtime/protocol.mjs

@@ -217,31 +217,6 @@ const normalizeOutputSchema = (value) => {
   return { name, schema: object.schema };
 };
 
-const toRunnerInput = (request, runId, models) => {
-  const normalized =
-    request?.model && request?.reasoningEffort && request?.version === GENERATION_VERSION
-      ? request
-      : normalizeGenerationRequest(request, models);
-  const input = normalized.input.map((message) => ({
-    role: message.role,
-    content: message.content.map((part) => {
-      if (part.type === 'text') return { type: 'text', text: part.text };
-      if (part.type === 'image') return { type: 'image', mimeType: part.mimeType, data: part.data };
-      throw unsupported(`input ${part.type}`);
-    }),
-  }));
-  return {
-    runId,
-    instructions: normalized.instructions,
-    messages: input,
-    output: normalized.output,
-    imageGeneration: normalized.output.includes('image'),
-    model: normalized.model,
-    reasoningEffort: normalized.reasoningEffort,
-    ...(normalized.outputSchema === undefined ? {} : { outputSchema: normalized.outputSchema }),
-  };
-};
-
 const normalizeGenerationRequest = (value, models) => {
   const request = asObject(value);
   if (request?.version !== GENERATION_VERSION) {
@@ -380,5 +355,4 @@ export {
   normalizeModelCatalog,
   publicModelCatalog,
   requestDigest,
-  toRunnerInput,
 };