@harness-engineering/orchestrator 0.11.1 → 0.11.2

package/dist/index.js

@@ -4084,7 +4084,7 @@ var LocalModelResolver = class {
 };
 
 // src/orchestrator.ts
-var import_local_models4 = require("@harness-engineering/local-models");
+var import_local_models5 = require("@harness-engineering/local-models");
 var import_core18 = require("@harness-engineering/core");
 
 // src/agent/config-migration.ts
@@ -9188,6 +9188,7 @@ async function runForceRefresh(res, scheduler, deps) {
 
 // src/server/routes/v1/local-models-pool-mutation.ts
 var import_core11 = require("@harness-engineering/core");
+var import_local_models3 = require("@harness-engineering/local-models");
 var INSTALL_RE = /^\/api\/v1\/local-models\/pool\/install(?:\?.*)?$/;
 var REMOVE_RE = /^\/api\/v1\/local-models\/pool\/remove(?:\?.*)?$/;
 var RESOLVE_TOP = 50;
@@ -9253,9 +9254,17 @@ async function handleInstall(req, res, deps) {
       evidence: match.evidence,
       freshness: `snapshot ${match.benchmarkSnapshot}`
     },
-    // 0 → the installer resolves the true on-disk size via `inspect` before the
-    // budget check, rather than trusting an estimate.
-    diskImpactGb: 0
+    // Estimate the on-disk size from the candidate's params + quant (the same
+    // `estimateDiskGb` the automated proposal engine uses). The pool needs a
+    // size for its pre-commit budget check *before* the pull; it cannot inspect
+    // the target first, because ollama `/api/show` 404s for a model that is not
+    // yet pulled locally — which would surface as a spurious "no longer
+    // available on HuggingFace" 404 on every operator install.
+    diskImpactGb: (0, import_local_models3.estimateDiskGb)({
+      sizeB: match.sizeB,
+      quant: match.quant,
+      ...match.activeB !== void 0 ? { activeB: match.activeB } : {}
+    })
   };
   const record = await (0, import_core11.createModelProposal)(deps.projectPath, content, {
     proposedBy: decidedByOf(deps, req)
@@ -14080,7 +14089,7 @@ var Orchestrator = class extends import_node_events.EventEmitter {
     if (overrides?.poolState) {
       this.poolStateProvider = overrides.poolState;
     } else if (localModelsEnabled) {
-      this.poolStateStore = new import_local_models4.PoolStateStore({
+      this.poolStateStore = new import_local_models5.PoolStateStore({
         onWarn: (message, cause) => this.logger.warn(message, cause !== void 0 ? { cause } : void 0)
       });
       this.poolStateProvider = this.poolStateStore;
@@ -15283,11 +15292,11 @@ ${messages}`);
   initModelPool(store) {
     const onWarn = (message, cause) => this.logger.warn(message, cause !== void 0 ? { cause } : void 0);
     const installerCfg = this.config.localModels?.installer;
-    this.modelInstaller = new import_local_models4.OllamaInstallAdapter({
+    this.modelInstaller = new import_local_models5.OllamaInstallAdapter({
       baseUrl: installerCfg?.ollamaEndpoint ?? "http://localhost:11434",
       onWarn
     });
-    this.modelPool = new import_local_models4.PoolManager({ store, installer: this.modelInstaller, onWarn });
+    this.modelPool = new import_local_models5.PoolManager({ store, installer: this.modelInstaller, onWarn });
   }
   /**
    * LMLM Phase 7 wiring: apply the operator's configured pool bounds (disk
@@ -15323,8 +15332,8 @@ ${messages}`);
     if (this.modelPool === null) return;
     const pool = this.modelPool;
     const refreshCfg = this.config.localModels?.refresh;
-    const frozen = (0, import_local_models4.loadFrozenCandidates)();
-    const candidates = (0, import_local_models4.selectCandidates)(frozen.candidates, this.config.localModels?.pool);
+    const frozen = (0, import_local_models5.loadFrozenCandidates)();
+    const candidates = (0, import_local_models5.selectCandidates)(frozen.candidates, this.config.localModels?.pool);
     for (const warning of frozen.warnings) {
       this.logger.warn("LMLM frozen candidate snapshot degraded", { warning });
     }
@@ -15334,10 +15343,10 @@ ${messages}`);
         allowedOrgs: this.config.localModels?.pool?.allowedOrgs ?? []
       });
     }
-    const recommend = (0, import_local_models4.createNativeRecommender)({ candidates });
+    const recommend = (0, import_local_models5.createNativeRecommender)({ candidates });
     this.modelRecommender = recommend;
-    this.refreshScheduler = new import_local_models4.RefreshScheduler({
-      runTick: () => (0, import_local_models4.runRefreshTick)({
+    this.refreshScheduler = new import_local_models5.RefreshScheduler({
+      runTick: () => (0, import_local_models5.runRefreshTick)({
         detectHardware: () => this.detectLmlmHardware(),
         recommend,
         poolManager: pool,
@@ -15368,7 +15377,7 @@ ${messages}`);
   /** Resolve the hardware profile for a refresh tick (operator override wins). */
   async detectLmlmHardware() {
     const override = this.config.localModels?.hardware?.override;
-    const detector = new import_local_models4.HardwareDetector(override !== void 0 ? { override } : {});
+    const detector = new import_local_models5.HardwareDetector(override !== void 0 ? { override } : {});
     return (await detector.detect()).profile;
   }
   /** Map the on-disk model-proposal queue to F7 dedup pairs (open→pending, rejected→rejected). */

package/dist/index.mjs

@@ -9143,6 +9143,7 @@ async function runForceRefresh(res, scheduler, deps) {
 
 // src/server/routes/v1/local-models-pool-mutation.ts
 import { createModelProposal, updateProposal as updateProposal4 } from "@harness-engineering/core";
+import { estimateDiskGb } from "@harness-engineering/local-models";
 var INSTALL_RE = /^\/api\/v1\/local-models\/pool\/install(?:\?.*)?$/;
 var REMOVE_RE = /^\/api\/v1\/local-models\/pool\/remove(?:\?.*)?$/;
 var RESOLVE_TOP = 50;
@@ -9208,9 +9209,17 @@ async function handleInstall(req, res, deps) {
       evidence: match.evidence,
       freshness: `snapshot ${match.benchmarkSnapshot}`
     },
-    // 0 → the installer resolves the true on-disk size via `inspect` before the
-    // budget check, rather than trusting an estimate.
-    diskImpactGb: 0
+    // Estimate the on-disk size from the candidate's params + quant (the same
+    // `estimateDiskGb` the automated proposal engine uses). The pool needs a
+    // size for its pre-commit budget check *before* the pull; it cannot inspect
+    // the target first, because ollama `/api/show` 404s for a model that is not
+    // yet pulled locally — which would surface as a spurious "no longer
+    // available on HuggingFace" 404 on every operator install.
+    diskImpactGb: estimateDiskGb({
+      sizeB: match.sizeB,
+      quant: match.quant,
+      ...match.activeB !== void 0 ? { activeB: match.activeB } : {}
+    })
   };
   const record = await createModelProposal(deps.projectPath, content, {
     proposedBy: decidedByOf(deps, req)

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@harness-engineering/orchestrator",
-  "version": "0.11.1",
+  "version": "0.11.2",
   "description": "Orchestrator daemon for dispatching coding agents to issues",
   "main": "./dist/index.js",
   "module": "./dist/index.mjs",
@@ -50,8 +50,8 @@
     "zod": "^3.25.76",
     "@harness-engineering/core": "0.34.1",
     "@harness-engineering/graph": "0.11.5",
-    "@harness-engineering/intelligence": "0.4.3",
     "@harness-engineering/local-models": "0.4.0",
+    "@harness-engineering/intelligence": "0.4.3",
     "@harness-engineering/types": "0.19.0"
   },
   "devDependencies": {