Adding label for deepinfra. Adding wrapper for claude integration with DeepInfra

.github/labeler.yml

@@ -325,3 +325,7 @@
   - changed-files:
       - any-glob-to-any-file:
           - "extensions/fal/**"
+"extensions: deepinfra":
+  - changed-files:
+      - any-glob-to-any-file:
+          - "extensions/deepinfra/**"

extensions/deepinfra/index.ts

@@ -1,6 +1,7 @@
 import { definePluginEntry } from "openclaw/plugin-sdk/core";
 import { createProviderApiKeyAuthMethod } from "openclaw/plugin-sdk/provider-auth";
 import { buildSingleProviderApiKeyCatalog } from "openclaw/plugin-sdk/provider-catalog";
+import { createDeepInfraSystemCacheWrapper } from "openclaw/plugin-sdk/provider-stream";
 import { applyDeepInfraConfig, DEEPINFRA_DEFAULT_MODEL_REF } from "./onboard.js";
 import { buildDeepInfraProviderWithDiscovery } from "./provider-catalog.js";
 
@@ -65,6 +66,7 @@ export default definePluginEntry({
         geminiThoughtSignatureModelHints: ["gemini"],
         dropThinkingBlockModelHints: ["claude"],
       },
+      wrapStreamFn: (ctx) => createDeepInfraSystemCacheWrapper(ctx.streamFn),
       isCacheTtlEligible: (ctx) => isDeepInfraCacheTtlModel(ctx.modelId),
     });
   },

src/agents/pi-embedded-runner/extra-params.deepinfra-cache-control.test.ts

@@ -0,0 +1,92 @@
+import type { Model } from "@mariozechner/pi-ai";
+import { describe, expect, it } from "vitest";
+import { runExtraParamsCase } from "./extra-params.test-support.js";
+
+type StreamPayload = {
+  messages: Array<{
+    role: string;
+    content: unknown;
+  }>;
+};
+
+function runDeepInfraPayload(payload: StreamPayload, modelId: string) {
+  runExtraParamsCase({
+    cfg: {
+      plugins: {
+        entries: {
+          deepinfra: {
+            enabled: true,
+          },
+        },
+      },
+    },
+    model: {
+      api: "openai-completions",
+      provider: "deepinfra",
+      id: modelId,
+    } as Model<"openai-completions">,
+    payload,
+  });
+}
+
+describe("extra-params: DeepInfra Anthropic cache_control", () => {
+  it("injects cache_control into system message for DeepInfra Anthropic models", () => {
+    const payload = {
+      messages: [
+        { role: "system", content: "You are a helpful assistant." },
+        { role: "user", content: "Hello" },
+      ],
+    };
+
+    runDeepInfraPayload(payload, "anthropic/claude-opus-4-6");
+
+    expect(payload.messages[0].content).toEqual([
+      { type: "text", text: "You are a helpful assistant.", cache_control: { type: "ephemeral" } },
+    ]);
+    expect(payload.messages[1].content).toBe("Hello");
+  });
+
+  it("adds cache_control to last content block when system message is already array", () => {
+    const payload = {
+      messages: [
+        {
+          role: "system",
+          content: [
+            { type: "text", text: "Part 1" },
+            { type: "text", text: "Part 2" },
+          ],
+        },
+      ],
+    };
+
+    runDeepInfraPayload(payload, "anthropic/claude-opus-4-6");
+
+    const content = payload.messages[0].content as Array<Record<string, unknown>>;
+    expect(content[0]).toEqual({ type: "text", text: "Part 1" });
+    expect(content[1]).toEqual({
+      type: "text",
+      text: "Part 2",
+      cache_control: { type: "ephemeral" },
+    });
+  });
+
+  it("does not inject cache_control for DeepInfra non-Anthropic models", () => {
+    const payload = {
+      messages: [{ role: "system", content: "You are a helpful assistant." }],
+    };
+
+    runDeepInfraPayload(payload, "google/gemini-2.5-pro");
+
+    expect(payload.messages[0].content).toBe("You are a helpful assistant.");
+  });
+
+  it("leaves payload unchanged when no system message exists", () => {
+    const payload = {
+      messages: [{ role: "user", content: "Hello" }],
+    };
+
+    runDeepInfraPayload(payload, "anthropic/claude-opus-4-6");
+
+    expect(payload.messages[0].content).toBe("Hello");
+  });
+});

src/agents/pi-embedded-runner/proxy-stream-wrappers.ts

@@ -15,6 +15,10 @@ function isOpenRouterAnthropicModel(provider: string, modelId: string): boolean
   return provider.toLowerCase() === "openrouter" && modelId.toLowerCase().startsWith("anthropic/");
 }
 
+function isDeepInfraAnthropicModel(provider: string, modelId: string): boolean {
+  return provider.toLowerCase() === "deepinfra" && modelId.toLowerCase().startsWith("anthropic/");
+}
+
 function mapThinkingLevelToOpenRouterReasoningEffort(
   thinkingLevel: ThinkLevel,
 ): "none" | "minimal" | "low" | "medium" | "high" | "xhigh" {
@@ -55,13 +59,16 @@ function normalizeProxyReasoningPayload(payload: unknown, thinkingLevel?: ThinkL
   }
 }
 
-export function createOpenRouterSystemCacheWrapper(baseStreamFn: StreamFn | undefined): StreamFn {
+export function createSystemCacheWrapper(
+  baseStreamFn: StreamFn | undefined,
+  isEligible: (provider: string, modelId: string) => boolean,
+): StreamFn {
   const underlying = baseStreamFn ?? streamSimple;
   return (model, context, options) => {
     if (
       typeof model.provider !== "string" ||
       typeof model.id !== "string" ||
-      !isOpenRouterAnthropicModel(model.provider, model.id)
+      !isEligible(model.provider, model.id)
     ) {
       return underlying(model, context, options);
     }
@@ -94,6 +101,14 @@ export function createOpenRouterSystemCacheWrapper(baseStreamFn: StreamFn | unde
   };
 }
 
+export function createOpenRouterSystemCacheWrapper(baseStreamFn: StreamFn | undefined): StreamFn {
+  return createSystemCacheWrapper(baseStreamFn, isOpenRouterAnthropicModel);
+}
+
+export function createDeepInfraSystemCacheWrapper(baseStreamFn: StreamFn | undefined): StreamFn {
+  return createSystemCacheWrapper(baseStreamFn, isDeepInfraAnthropicModel);
+}
+
 export function createOpenRouterWrapper(
   baseStreamFn: StreamFn | undefined,
   thinkingLevel?: ThinkLevel,

src/plugin-sdk/provider-stream.ts

@@ -9,9 +9,11 @@ export {
   sanitizeGoogleThinkingPayload,
 } from "../agents/pi-embedded-runner/google-stream-wrappers.js";
 export {
+  createDeepInfraSystemCacheWrapper,
   createKilocodeWrapper,
   createOpenRouterSystemCacheWrapper,
   createOpenRouterWrapper,
+  createSystemCacheWrapper,
   isProxyReasoningUnsupported,
 } from "../agents/pi-embedded-runner/proxy-stream-wrappers.js";
 export {