NVIDIA · ericksoa · Aug 4, 2026 · Aug 3, 2026 · Aug 3, 2026 · Aug 3, 2026
diff --git a/managed-inference/schemas/catalog.schema.json b/managed-inference/schemas/catalog.schema.json
@@ -0,0 +1,53 @@
+{
+  "$schema": "https://json-schema.org/draft/2020-12/schema",
+  "$id": "https://github.com/NVIDIA/NemoClaw/managed-inference/schemas/catalog.schema.json",
+  "$comment": "SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.\nSPDX-License-Identifier: Apache-2.0",
+  "title": "NemoClaw compiled managed inference serving catalog",
+  "type": "object",
+  "required": [
+    "schemaVersion",
+    "compilerVersion",
+    "sourceRevision",
+    "readinessSchemaRef",
+    "recipes",
+    "presets",
+    "sources",
+    "catalogDigest"
+  ],
+  "properties": {
+    "schemaVersion": { "const": "1.0.0" },
+    "compilerVersion": { "const": "1.0.0" },
+    "sourceRevision": { "type": "string", "pattern": "^[0-9a-f]{40,64}$" },
+    "readinessSchemaRef": {
+      "const": "https://github.com/NVIDIA/NemoClaw/schemas/system-readiness.schema.json"
+    },
+    "recipes": {
+      "type": "array",
+      "items": { "$ref": "recipe.schema.json" }
+    },
+    "presets": {
+      "type": "array",
+      "items": { "$ref": "preset.schema.json" }
+    },
+    "sources": {
+      "type": "array",
+      "items": {
+        "type": "object",
+        "required": ["path", "kind", "id", "digest"],
+        "properties": {
+          "path": {
+            "type": "string",
+            "pattern": "^managed-inference/(?:recipes|presets)/.+\\.ya?ml$",
+            "maxLength": 512
+          },
+          "kind": { "enum": ["ServingRecipe", "ServingPreset"] },
+          "id": { "type": "string", "minLength": 1, "maxLength": 160 },
+          "digest": { "type": "string", "pattern": "^sha256:[0-9a-f]{64}$" }
+        },
+        "additionalProperties": false
+      }
+    },
+    "catalogDigest": { "type": "string", "pattern": "^sha256:[0-9a-f]{64}$" }
+  },
+  "additionalProperties": false
+}
diff --git a/managed-inference/schemas/preset.schema.json b/managed-inference/schemas/preset.schema.json
@@ -0,0 +1,96 @@
+{
+  "$schema": "https://json-schema.org/draft/2020-12/schema",
+  "$id": "https://github.com/NVIDIA/NemoClaw/managed-inference/schemas/preset.schema.json",
+  "$comment": "SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.\nSPDX-License-Identifier: Apache-2.0",
+  "title": "NemoClaw managed inference serving preset",
+  "type": "object",
+  "required": ["apiVersion", "kind", "metadata", "spec"],
+  "properties": {
+    "apiVersion": { "const": "nemoclaw.nvidia.com/managed-inference/v1" },
+    "kind": { "const": "ServingPreset" },
+    "metadata": {
+      "type": "object",
+      "required": ["id"],
+      "properties": {
+        "id": { "$ref": "#/$defs/stableId" },
+        "displayName": { "type": "string", "minLength": 1, "maxLength": 256 }
+      },
+      "additionalProperties": false
+    },
+    "spec": {
+      "type": "object",
+      "required": ["selection", "priority", "plan"],
+      "properties": {
+        "selection": { "enum": ["automatic", "explicit-only", "disabled"] },
+        "priority": { "type": "integer", "minimum": 0, "maximum": 10000 },
+        "requirements": {
+          "type": "object",
+          "required": ["all"],
+          "properties": {
+            "all": {
+              "type": "array",
+              "maxItems": 64,
+              "items": { "$ref": "#/$defs/readinessRequirement" }
+            }
+          },
+          "additionalProperties": false
+        },
+        "plan": {
+          "type": "object",
+          "required": ["backend", "recipeRef"],
+          "properties": {
+            "backend": { "$ref": "#/$defs/backendId" },
+            "recipeRef": { "$ref": "#/$defs/stableId" }
+          },
+          "additionalProperties": false
+        }
+      },
+      "additionalProperties": false
+    }
+  },
+  "additionalProperties": false,
+  "$defs": {
+    "stableId": {
+      "type": "string",
+      "pattern": "^[a-z][a-z0-9]*(?:[._/-][a-z0-9]+)+$",
+      "maxLength": 160
+    },
+    "backendId": {
+      "type": "string",
+      "pattern": "^[a-z][a-z0-9-]{0,63}$"
+    },
+    "readinessRequirement": {
+      "type": "object",
+      "required": ["readiness"],
+      "properties": {
+        "readiness": {
+          "oneOf": [
+            {
+              "type": "object",
+              "required": ["scope", "kind", "id", "state"],
+              "properties": {
+                "scope": { "enum": ["controller", "everyNode"] },
+                "kind": { "enum": ["observation", "capability"] },
+                "id": { "$ref": "#/$defs/stableId" },
+                "state": { "enum": ["present", "absent"] }
+              },
+              "additionalProperties": false
+            },
+            {
+              "type": "object",
+              "required": ["scope", "kind", "id", "status"],
+              "properties": {
+                "scope": { "enum": ["controller", "everyNode"] },
+                "kind": { "const": "qualification" },
+                "id": { "$ref": "#/$defs/stableId" },
+                "status": { "enum": ["qualified", "unqualified"] }
+              },
+              "additionalProperties": false
+            }
+          ]
+        }
+      },
+      "additionalProperties": false
+    }
+  }
+}
diff --git a/managed-inference/schemas/recipe.schema.json b/managed-inference/schemas/recipe.schema.json
@@ -0,0 +1,151 @@
+{
+  "$schema": "https://json-schema.org/draft/2020-12/schema",
+  "$id": "https://github.com/NVIDIA/NemoClaw/managed-inference/schemas/recipe.schema.json",
+  "$comment": "SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.\nSPDX-License-Identifier: Apache-2.0",
+  "title": "NemoClaw managed inference serving recipe",
+  "type": "object",
+  "required": ["apiVersion", "kind", "metadata", "spec"],
+  "properties": {
+    "apiVersion": { "const": "nemoclaw.nvidia.com/managed-inference/v1" },
+    "kind": { "const": "ServingRecipe" },
+    "metadata": { "$ref": "#/$defs/metadata" },
+    "spec": {
+      "type": "object",
+      "required": ["backend", "model", "execution"],
+      "properties": {
+        "backend": { "$ref": "#/$defs/backendId" },
+        "model": {
+          "type": "object",
+          "required": ["id", "revision"],
+          "properties": {
+            "id": { "type": "string", "minLength": 1, "maxLength": 256 },
+            "revision": {
+              "type": "string",
+              "pattern": "^(?:[0-9a-f]{40}|[0-9a-f]{64}|sha256:[0-9a-f]{64})$"
+            },
+            "servedName": { "$ref": "#/$defs/stableId" },
+            "files": {
+              "type": "array",
+              "minItems": 1,
+              "maxItems": 32,
+              "items": {
+                "type": "object",
+                "required": ["path", "digest"],
+                "properties": {
+                  "path": {
+                    "type": "string",
+                    "minLength": 1,
+                    "maxLength": 512,
+                    "pattern": "^(?!\\.{1,2}(?:/|$))(?!.*\\/\\.{1,2}(?:/|$))[A-Za-z0-9._-]+(?:/[A-Za-z0-9._-]+)*$"
+                  },
+                  "digest": { "$ref": "#/$defs/sha256" }
+                },
+                "additionalProperties": false
+              }
+            }
+          },
+          "additionalProperties": false
+        },
+        "runtime": {
+          "type": "object",
+          "properties": {
+            "image": {
+              "type": "string",
+              "pattern": "^(?:[a-z0-9]+(?:[._-][a-z0-9]+)*(?::[0-9]+)?/)?(?:[a-z0-9]+(?:[._-][a-z0-9]+)*/)*[a-z0-9]+(?:[._-][a-z0-9]+)*@sha256:[0-9a-f]{64}$"
+            },
+            "architecture": { "enum": ["amd64", "arm64"] },
+            "components": {
+              "type": "object",
+              "maxProperties": 32,
+              "propertyNames": { "$ref": "#/$defs/stableId" },
+              "additionalProperties": {
+                "type": "string",
+                "minLength": 1,
+                "maxLength": 128
+              }
+            }
+          },
+          "additionalProperties": false
+        },
+        "execution": {
+          "type": "object",
+          "required": ["materializerRef", "lifecycleRef"],
+          "properties": {
+            "materializerRef": { "$ref": "#/$defs/adapterId" },
+            "lifecycleRef": { "$ref": "#/$defs/adapterId" }
+          },
+          "additionalProperties": false
+        },
+        "serve": {
+          "type": "object",
+          "properties": {
+            "arguments": {
+              "type": "array",
+              "maxItems": 128,
+              "items": {
+                "type": "object",
+                "required": ["name"],
+                "properties": {
+                  "name": {
+                    "type": "string",
+                    "pattern": "^--[a-z0-9][a-z0-9-]*$",
+                    "maxLength": 128
+                  },
+                  "value": { "$ref": "#/$defs/scalar" }
+                },
+                "additionalProperties": false
+              }
+            }
+          },
+          "additionalProperties": false
+        },
+        "readiness": {
+          "type": "object",
+          "properties": {
+            "timeoutSeconds": { "type": "integer", "minimum": 1, "maximum": 86400 },
+            "expectedModel": { "$ref": "#/$defs/stableId" }
+          },
+          "additionalProperties": false
+        }
+      },
+      "additionalProperties": false
+    }
+  },
+  "additionalProperties": false,
+  "$defs": {
+    "metadata": {
+      "type": "object",
+      "required": ["id"],
+      "properties": {
+        "id": { "$ref": "#/$defs/stableId" },
+        "displayName": { "type": "string", "minLength": 1, "maxLength": 256 }
+      },
+      "additionalProperties": false
+    },
+    "stableId": {
+      "type": "string",
+      "pattern": "^[a-z][a-z0-9]*(?:[._/-][a-z0-9]+)+$",
+      "maxLength": 160
+    },
+    "backendId": {
+      "type": "string",
+      "pattern": "^[a-z][a-z0-9-]{0,63}$"
+    },
+    "adapterId": {
+      "type": "string",
+      "pattern": "^[a-z][a-z0-9]*(?:[._-][a-z0-9]+)+/v[1-9][0-9]*$",
+      "maxLength": 160
+    },
+    "sha256": {
+      "type": "string",
+      "pattern": "^sha256:[0-9a-f]{64}$"
+    },
+    "scalar": {
+      "oneOf": [
+        { "type": "string", "maxLength": 1024 },
+        { "type": "number" },
+        { "type": "boolean" }
+      ]
+    }
+  }
+}
diff --git a/package.json b/package.json
@@ -55,7 +55,7 @@
     "check:installer-hash": "bash scripts/check-installer-hash.sh",
     "typecheck": "tsc -p jsconfig.json",
     "build:policy-boundary": "tsc -p nemoclaw/tsconfig.shared.json",
-    "build:cli": "npm run build:policy-boundary && tsc -p tsconfig.src.json && node dist/lib/core/generate-build-identity.js && node dist/lib/cli/generate-oclif-metadata-manifest.js && if find nemoclaw-blueprint/scripts -name '*.ts' -print -quit | grep -q .; then tsc -p nemoclaw-blueprint/tsconfig.json; fi",
+    "build:cli": "npm run build:policy-boundary && tsc -p tsconfig.src.json && node dist/lib/core/generate-build-identity.js && node dist/lib/inference/serving/generate-catalog.js && node dist/lib/cli/generate-oclif-metadata-manifest.js && if find nemoclaw-blueprint/scripts -name '*.ts' -print -quit | grep -q .; then tsc -p nemoclaw-blueprint/tsconfig.json; fi",
     "clean:cli": "node -e \"require('node:fs').rmSync('dist', { recursive: true, force: true })\"",
     "typecheck:cli": "tsc -p tsconfig.cli.json",
     "validate:configs": "tsx scripts/validate-configs.mts",
@@ -115,6 +115,7 @@
     "nemoclaw/openclaw.plugin.json",
     "nemoclaw/package.json",
     "nemoclaw-blueprint/",
+    "managed-inference/",
     "schemas/network-policy.schema.json",
     "schemas/sandbox-policy.schema.json",
     "scripts/",