pi-provider-litellm 3.2.0 → 3.4.0

package/src/types.ts

@@ -1,5 +1,6 @@
 import type { Model } from "@earendil-works/pi-ai";
 import type { BackendFamily, LITELLM_DISCOVERY_VERSION } from "./backend-identity.js";
+import type { ProxyVersion } from "./proxy-version.js";
 
 export type DiscoverySource = "model_info" | "models_list" | "health";
 
@@ -27,6 +28,14 @@ export interface LiteLLMModelPolicy {
   // Gemini-backed deployments require lowercase effort values. Discovery carries
   // this evidence so the request hook does not infer a backend from the route name.
   normalizeGeminiReasoningEffort?: boolean;
+  // GPT-5.5 and later reject `reasoning_effort` alongside function tools on
+  // /v1/chat/completions. Discovery sets this for Chat routes from deployment
+  // evidence so tool requests drop reasoning instead of failing over.
+  dropToolReasoning?: boolean;
+  // GPT-6 falls back to its default effort when `reasoning_effort` is omitted, and
+  // that is rejected alongside tools too, so these tool requests send `none` unless
+  // the model declares its own off effort.
+  explicitToolReasoningOff?: boolean;
 }
 
 export type DiscoveredModelFor<TApi extends LiteLLMApi> = Omit<Model<TApi>, "provider" | "baseUrl"> & {
@@ -56,6 +65,9 @@ export type LiteLLMModel = Model<LiteLLMApi> & {
 export interface DiscoveryResult {
   models: DiscoveredModel[];
   source: DiscoverySource;
+  // Present only when a `/model/info` discovery had to read it and the proxy
+  // answered. Absent means unknown, not old: discovery asks only when it must.
+  proxyVersion?: ProxyVersion;
 }
 
 export interface DiscoveryOptions {
@@ -82,7 +94,7 @@ export interface ModelInfoEntry {
     mode?: string | null;
     base_model?: string;
     litellm_provider?: string;
-    supported_endpoints?: string[];
+    supported_endpoints?: string[] | null;
     supported_openai_params?: string[];
     input_cost_per_token?: number;
     output_cost_per_token?: number;
@@ -138,6 +150,14 @@ export interface ResolvedCredentials {
   apiKeyFromGcloudAdc?: boolean;
 }
 
+// The boolean MCP tool annotation hints, with the meaning Pi gives `ToolAnnotations`.
+export interface McpToolAnnotations {
+  readOnlyHint?: boolean;
+  destructiveHint?: boolean;
+  idempotentHint?: boolean;
+  openWorldHint?: boolean;
+}
+
 export interface LiteLLMMcpTool {
   name: string;
   server_name: string;
@@ -147,6 +167,10 @@ export interface LiteLLMMcpTool {
   input_schema: Record<string, unknown>;
   // True when the proxy supplied an `inputSchema`/`input_schema` that was not a JSON object.
   input_schema_malformed?: boolean;
+  // Boolean hints from the proxy-supplied `annotations`; every other annotation field is dropped.
+  annotations?: McpToolAnnotations;
+  // `mcp_info.description` of the server, when the proxy supplies one.
+  server_description?: string;
 }
 
 export interface LiteLLMSkill {