Configure per-model context and output limits

contracts/gateway/v4/compatibility-manifest.generated.json

@@ -29,7 +29,7 @@
       "delivery": "live-only-best-effort",
       "family": "conversation.events",
       "schema": "conversation/conversation-events.schema.json",
-      "schemaSha256": "3a853963ea7b33c87edef960338dd1e181f6e5140a49762a9e345fea8b6bc37c",
+      "schemaSha256": "00a95e7de26a01838bfb2db882f092b9e499de19052aa1010e8a03b1adbdca8a",
       "schemaVersion": 1,
       "wireNames": [
         "chat.done",
@@ -724,6 +724,12 @@
       "schema": "platform/migration-sources-preview.schema.json",
       "schemaSha256": "7d89ea10bff333c3a5c92fb157ea887fbc5823685808e6531fab1cb9426597ff"
     },
+    {
+      "lifecycle": "stable",
+      "name": "models.capacity.resolve",
+      "schema": "platform/models-capacity-resolve.schema.json",
+      "schemaSha256": "ad2449d34174321d078ee87779b13457c998056897de6008ac9a41e8829851e0"
+    },
     {
       "lifecycle": "stable",
       "name": "models.list",
@@ -1463,10 +1469,10 @@
   "protocol": "opensquilla-websocket-json",
   "source": {
     "eventFamilyCount": 10,
-    "generatorSha256": "f781c6d8e31b336c2d2e342784935b3d327c9947cb35e56de8e44a0a83bffae6",
-    "methodCount": 215,
-    "schemaCount": 225,
-    "schemaTreeSha256": "840fa1aabc2cf9ec62fb500bbe7771c49a00a5f54e367cabc3865fa57fb63579"
+    "generatorSha256": "d2ba9f83594c975b279a95386cd247d50b6704d4b6feabd516c8d870ad827013",
+    "methodCount": 216,
+    "schemaCount": 226,
+    "schemaTreeSha256": "acfc84425d8709c6c39362c0cce19ae186462d430a1205ef43042778b6ff983b"
   },
   "wireVersion": 4
 }

contracts/gateway/v4/conversation/conversation-events.schema.json

@@ -147,6 +147,43 @@
       "type": "object",
       "additionalProperties": true,
       "properties": {
+        "model_capacity": {
+          "type": [
+            "object",
+            "null"
+          ],
+          "additionalProperties": false,
+          "required": [
+            "provider",
+            "model",
+            "contextWindow",
+            "source"
+          ],
+          "properties": {
+            "provider": {
+              "type": "string",
+              "minLength": 1,
+              "maxLength": 1024
+            },
+            "model": {
+              "type": "string",
+              "minLength": 1,
+              "maxLength": 1024
+            },
+            "contextWindow": {
+              "type": "integer",
+              "minimum": 1
+            },
+            "source": {
+              "enum": [
+                "default",
+                "catalog",
+                "config",
+                "override"
+              ]
+            }
+          }
+        },
         "key": {
           "type": ["string", "null"],
           "minLength": 1

contracts/gateway/v4/platform/models-capacity-resolve.schema.json

@@ -0,0 +1,256 @@
+{
+  "$schema": "https://json-schema.org/draft/2020-12/schema",
+  "$id": "https://opensquilla.dev/contracts/gateway/v4/platform/models-capacity-resolve.schema.json",
+  "title": "OpenSquilla models.capacity.resolve Contract",
+  "type": "object",
+  "additionalProperties": false,
+  "properties": {
+    "request": {
+      "$ref": "#/$defs/RequestFrame"
+    },
+    "response": {
+      "$ref": "#/$defs/ResponseFrame"
+    }
+  },
+  "x-opensquilla-wire": {
+    "protocol": "opensquilla-websocket-json",
+    "version": 4,
+    "compatibility": "exact-json-tree"
+  },
+  "x-opensquilla-codegen": {
+    "python": {
+      "tool": "datamodel-code-generator",
+      "version": "0.75.1",
+      "target": "pydantic_v2.BaseModel"
+    },
+    "typescript": {
+      "tool": "json-schema-to-typescript",
+      "version": "15.0.4"
+    },
+    "runtimeValidation": {
+      "tool": "ajv",
+      "version": "8.17.1",
+      "mode": "standalone-adapter-only"
+    }
+  },
+  "x-opensquilla-method": {
+    "name": "models.capacity.resolve",
+    "kind": "query",
+    "scope": "operator.read",
+    "guestAllowed": false,
+    "idempotency": "read-only",
+    "timeout": {
+      "policy": "transport"
+    },
+    "capability": {
+      "kind": "method-availability",
+      "name": "models.capacity.resolve"
+    },
+    "request": "#/$defs/RequestFrame",
+    "params": "#/$defs/Params",
+    "response": "#/$defs/ResponseFrame",
+    "result": "#/$defs/Result",
+    "errors": [
+      {
+        "code": "INVALID_REQUEST"
+      },
+      {
+        "code": "UNAUTHORIZED"
+      },
+      {
+        "code": "UNAVAILABLE",
+        "retryable": true
+      },
+      {
+        "code": "INTERNAL_ERROR"
+      }
+    ]
+  },
+  "$defs": {
+    "Params": {
+      "type": "object",
+      "additionalProperties": false,
+      "required": [
+        "models"
+      ],
+      "properties": {
+        "models": {
+          "type": "array",
+          "maxItems": 128,
+          "items": {
+            "$ref": "#/$defs/Target"
+          }
+        }
+      }
+    },
+    "Result": {
+      "type": "object",
+      "additionalProperties": false,
+      "required": [
+        "models"
+      ],
+      "properties": {
+        "models": {
+          "type": "array",
+          "items": {
+            "$ref": "#/$defs/Capacity"
+          }
+        }
+      }
+    },
+    "RpcError": {
+      "type": "object",
+      "additionalProperties": true
+    },
+    "RequestFrame": {
+      "type": "object",
+      "additionalProperties": true,
+      "required": [
+        "type",
+        "id",
+        "method"
+      ],
+      "properties": {
+        "type": {
+          "const": "req"
+        },
+        "id": {
+          "type": "string"
+        },
+        "method": {
+          "const": "models.capacity.resolve"
+        }
+      }
+    },
+    "ResponseFrame": {
+      "type": "object",
+      "additionalProperties": true,
+      "required": [
+        "type",
+        "id",
+        "ok"
+      ],
+      "properties": {
+        "type": {
+          "const": "res"
+        },
+        "id": {
+          "type": "string"
+        },
+        "ok": {
+          "type": "boolean"
+        },
+        "payload": {
+          "$ref": "#/$defs/Result"
+        },
+        "error": {
+          "$ref": "#/$defs/RpcError"
+        }
+      }
+    },
+    "Target": {
+      "type": "object",
+      "additionalProperties": false,
+      "required": [
+        "provider",
+        "model"
+      ],
+      "properties": {
+        "provider": {
+          "type": "string",
+          "minLength": 1,
+          "maxLength": 1024,
+          "pattern": "\\S"
+        },
+        "model": {
+          "type": "string",
+          "minLength": 1,
+          "maxLength": 1024,
+          "pattern": "\\S"
+        }
+      }
+    },
+    "Limit": {
+      "type": "object",
+      "additionalProperties": false,
+      "required": [
+        "automatic",
+        "automaticSource",
+        "override",
+        "value",
+        "source",
+        "editable"
+      ],
+      "properties": {
+        "automatic": {
+          "type": "integer",
+          "minimum": 1
+        },
+        "automaticSource": {
+          "type": "string",
+          "enum": [
+            "catalog",
+            "default",
+            "override",
+            "config"
+          ]
+        },
+        "override": {
+          "type": [
+            "integer",
+            "null"
+          ],
+          "minimum": 1
+        },
+        "value": {
+          "type": "integer",
+          "minimum": 1
+        },
+        "source": {
+          "type": "string",
+          "enum": [
+            "catalog",
+            "default",
+            "override",
+            "config"
+          ]
+        },
+        "editable": {
+          "type": "boolean"
+        }
+      }
+    },
+    "Capacity": {
+      "type": "object",
+      "additionalProperties": false,
+      "required": [
+        "provider",
+        "model",
+        "contextWindow",
+        "maxOutputTokens",
+        "localRuntime"
+      ],
+      "properties": {
+        "provider": {
+          "type": "string",
+          "minLength": 1,
+          "maxLength": 1024
+        },
+        "model": {
+          "type": "string",
+          "minLength": 1,
+          "maxLength": 1024
+        },
+        "contextWindow": {
+          "$ref": "#/$defs/Limit"
+        },
+        "maxOutputTokens": {
+          "$ref": "#/$defs/Limit"
+        },
+        "localRuntime": {
+          "type": "boolean"
+        }
+      }
+    }
+  }
+}

contracts/gateway/v4/production-targets.json

@@ -4,27 +4,40 @@
     {
       "kind": "method",
       "wireName": "telemetry.product_active.record",
-      "roles": ["result"]
+      "roles": [
+        "result"
+      ]
     },
     {
       "kind": "method",
       "wireName": "sessions.executionLog.read",
-      "roles": ["params", "result"]
+      "roles": [
+        "params",
+        "result"
+      ]
     },
     {
       "kind": "event",
       "wireName": "transport.flow.dirty",
-      "roles": ["payload"]
+      "roles": [
+        "payload"
+      ]
     },
     {
       "kind": "method",
       "wireName": "transport.flow.update",
-      "roles": ["params", "result"]
+      "roles": [
+        "params",
+        "result"
+      ]
     },
     {
       "kind": "method",
       "wireName": "sessions.messages.snapshot.read",
-      "roles": ["params", "result"]
+      "roles": [
+        "params",
+        "result"
+      ]
     },
     {
       "kind": "event",
@@ -678,6 +691,14 @@
         "result"
       ]
     },
+    {
+      "kind": "method",
+      "wireName": "models.capacity.resolve",
+      "roles": [
+        "params",
+        "result"
+      ]
+    },
     {
       "kind": "method",
       "wireName": "models.list",