fix(deepseek): backfill v4 reasoning for proxy models

steipete · steipete · commit b97cb15b078f · 2026-05-10T05:48:02.000+01:00
diff --git a/CHANGELOG.md b/CHANGELOG.md
@@ -61,6 +61,7 @@ Docs: https://docs.openclaw.ai
 - Channels/iMessage: wire `action: "reply"` attachments through `imsg send-rich --file` when the installed imsg build advertises that capability (probed once via `imsg send-rich --help` and cached on the private-API status). Reply now hydrates `media`/`mediaUrl`/`fileUrl`/`mediaUrls[0]`/`filePath`/`path`/base64 `buffer`+`filename` through the shared outbound resolver, stages buffers via the existing `withTempFile` helper, rejects `http(s)://` URL attachments with a targeted error pointing callers at `send`'s full attachment-resolver pipeline, and falls back to the explicit `imsg#114 not landed yet` error on older imsg builds. Depends on the upstream `openclaw/imsg#114` capability landing in an installable release; until then the new path stays gated and users see the same explicit fallback `#79822` introduced. (#79864) Thanks @omarshahine.
 - Telegram: preserve the first-preview debounce while appending true partial-stream deltas, so edited draft previews no longer duplicate earlier text when providers emit incremental output. (#80045) Thanks @TurboTheTurtle.
 - Volcengine/Kimi: strip provider-unsupported tool schema length and item constraint keywords for direct and coding-plan models so hosted Kimi runs do not reject message tools with `minLength`. Fixes #38817.
+- DeepSeek: backfill V4 `reasoning_content` replay fields for unowned OpenAI-compatible proxy providers, preventing follow-up request failures outside the bundled DeepSeek and OpenRouter routes. Fixes #79608.
 
 ## 2026.5.9
 
diff --git a/src/agents/pi-embedded-runner-extraparams.test.ts b/src/agents/pi-embedded-runner-extraparams.test.ts
@@ -678,6 +678,33 @@ describe("applyExtraParamsToAgent", () => {
     expect(payloads[0]?.thinking).toEqual({ type: "disabled" });
   });
 
+  it("fills DeepSeek V4 reasoning_content for unowned OpenAI-compatible proxy models", () => {
+    const payload = runResponsesPayloadMutationCase({
+      applyProvider: "opencode",
+      applyModelId: "deepseek-v4-pro",
+      thinkingLevel: "high",
+      model: {
+        api: "openai-completions",
+        provider: "opencode",
+        id: "deepseek-v4-pro",
+      } as Model<"openai-completions">,
+      payload: {
+        messages: [
+          { role: "user", content: "continue" },
+          { role: "assistant", content: "I used a tool" },
+          { role: "tool", content: "ok" },
+        ],
+      },
+    });
+
+    const messages = payload.messages as Array<Record<string, unknown>>;
+    expect(payload.thinking).toEqual({ type: "enabled" });
+    expect(payload.reasoning_effort).toBe("high");
+    expect(messages[0]).not.toHaveProperty("reasoning_content");
+    expect(messages[1]).toHaveProperty("reasoning_content", "");
+    expect(messages[2]).not.toHaveProperty("reasoning_content");
+  });
+
   it("strips xai Responses reasoning payload fields", () => {
     const payload = runResponsesPayloadMutationCase({
       applyProvider: "xai",
diff --git a/src/agents/pi-embedded-runner/extra-params.ts b/src/agents/pi-embedded-runner/extra-params.ts
@@ -4,6 +4,7 @@ import { streamSimple } from "@mariozechner/pi-ai";
 import type { SettingsManager } from "@mariozechner/pi-coding-agent";
 import type { ThinkLevel } from "../../auto-reply/thinking.js";
 import type { OpenClawConfig } from "../../config/types.openclaw.js";
+import { createDeepSeekV4OpenAICompatibleThinkingWrapper } from "../../plugin-sdk/provider-stream-shared.js";
 import {
   prepareProviderExtraParams as prepareProviderExtraParamsRuntime,
   type ProviderRuntimePluginHandle,
@@ -689,6 +690,12 @@ function applyPostPluginStreamWrappers(
   ctx.agent.streamFn = createOpenAICompletionsToolsCompatWrapper(ctx.agent.streamFn);
 
   if (!ctx.providerWrapperHandled) {
+    ctx.agent.streamFn = createDeepSeekV4OpenAICompatibleThinkingWrapper({
+      baseStreamFn: ctx.agent.streamFn,
+      thinkingLevel: ctx.thinkingLevel,
+      shouldPatchModel: isDeepSeekV4OpenAICompatibleModel,
+    });
+
     // Guard Google-family payloads against invalid negative thinking budgets
     // emitted by upstream model-ID heuristics for Gemini 3.1 variants.
     ctx.agent.streamFn = createGoogleThinkingPayloadWrapper(ctx.agent.streamFn, ctx.thinkingLevel);
@@ -752,6 +759,22 @@ function applyPostPluginStreamWrappers(
   log.warn(`ignoring invalid parallel_tool_calls param: ${summary}`);
 }
 
+function normalizeDeepSeekV4CandidateId(modelId: unknown): string | undefined {
+  if (typeof modelId !== "string") {
+    return undefined;
+  }
+  const withoutSuffix = modelId.trim().toLowerCase().split(":", 1)[0];
+  return withoutSuffix.split("/").pop();
+}
+
+function isDeepSeekV4OpenAICompatibleModel(model: Parameters<StreamFn>[0]): boolean {
+  const normalizedModelId = normalizeDeepSeekV4CandidateId(model.id);
+  return (
+    model.api === "openai-completions" &&
+    (normalizedModelId === "deepseek-v4-flash" || normalizedModelId === "deepseek-v4-pro")
+  );
+}
+
 /**
  * Apply extra params (like temperature) to an agent's streamFn.
  * Also applies verified provider-specific request wrappers, such as OpenRouter attribution.
