diff --git a/changelog.d/fixes/responses-additional-tools.md b/changelog.d/fixes/responses-additional-tools.md
new file mode 100644
index 0000000000..a340625de7
--- /dev/null
+++ b/changelog.d/fixes/responses-additional-tools.md
@@ -0,0 +1 @@
+- **fix(responses):** Make Responses API to Chat Completions downgrades provider-neutral and loss-aware. Tools declared through `additional_tools` are preserved, structured output and `allowed_tools` are translated, refusal history is normalized, unsupported input item types fail explicitly instead of disappearing, and Responses-only execution/cache fields no longer leak to strict Chat endpoints.
diff --git a/open-sse/translator/request/openai-responses.ts b/open-sse/translator/request/openai-responses.ts
index 7141d9f946..b6d7679cc6 100644
--- a/open-sse/translator/request/openai-responses.ts
+++ b/open-sse/translator/request/openai-responses.ts
@@ -8,6 +8,7 @@ import { isOpenAIResponsesStoreEnabled } from "@/lib/providers/requestDefaults";
 import { FORMATS } from "../formats.ts";
 import { register } from "../registry.ts";
 import { normalizeResponsesInputForChat } from "../../utils/responsesInputNormalization.ts";
+import { collectResponsesTools } from "./openai-responses/additionalTools.ts";
 import { openaiToOpenAIResponsesRequest } from "./openai-responses/toResponses.ts";
 import {
   JsonRecord,
@@ -17,7 +18,6 @@ import {
   TOOL_SEARCH_TOOL_TYPES,
   IMAGE_GENERATION_TOOL_TYPES,
   toRecord,
-  toArray,
   toString,
   normalizeVerbosity,
   normalizeResponsesReasoningEffort,
@@ -46,9 +46,12 @@ export function openaiResponsesToOpenAIRequest(
   if (root.input === undefined) return body;
   const credentialRecord = toRecord(credentials);
   const storeEnabled = isOpenAIResponsesStoreEnabled(credentialRecord.providerSpecificData);
+  const rawInputItems = normalizeResponsesInputForChat(root.input);
 
-  // Validate tool types — only function tools can be translated to Chat Completions
-  const tools = toArray(root.tools);
+  // Tools may be declared at the Responses top level or in one or more
+  // `additional_tools` input items. Normalize both forms before validation/conversion so
+  // every downgraded provider receives the same available tool set.
+  const tools = collectResponsesTools(root.tools, rawInputItems);
   if (tools.length > 0) {
     for (const toolValue of tools) {
       const tool = toRecord(toolValue);
@@ -101,6 +104,20 @@ export function openaiResponsesToOpenAIRequest(
   // `text` object (a strict Chat endpoint 400s on unknown fields).
   const responsesVerbosity = normalizeVerbosity(toRecord(result.text).verbosity);
   if (responsesVerbosity && isOpenAIDestination) result.verbosity = responsesVerbosity;
+  const responsesTextFormat = toRecord(toRecord(result.text).format);
+  if (responsesTextFormat.type === "json_schema" && responsesTextFormat.schema !== undefined) {
+    const jsonSchema: JsonRecord = {
+      name: toString(responsesTextFormat.name, "response"),
+      schema: responsesTextFormat.schema,
+    };
+    if (responsesTextFormat.description !== undefined) {
+      jsonSchema.description = responsesTextFormat.description;
+    }
+    if (responsesTextFormat.strict !== undefined) jsonSchema.strict = responsesTextFormat.strict;
+    result.response_format = { type: "json_schema", json_schema: jsonSchema };
+  } else if (responsesTextFormat.type === "json_object") {
+    result.response_format = { type: "json_object" };
+  }
   delete result.text;
 
   // background: true requests a deferred Responses API run (the upstream
@@ -138,7 +155,6 @@ export function openaiResponsesToOpenAIRequest(
   // Upstream providers reject messages:[] with "400: at least one message is required".
   // When the client sends input:[] (empty), inject a placeholder user message — mirrors
   // upstream 9router#419 (and the existing empty-string handling elsewhere in this file).
-  const rawInputItems = normalizeResponsesInputForChat(root.input);
   const inputItems: unknown[] =
     rawInputItems.length === 0
       ? [{ type: "message", role: "user", content: [{ type: "input_text", text: "..." }] }]
@@ -175,6 +191,9 @@ export function openaiResponsesToOpenAIRequest(
             if (contentItem.type === "output_text") {
               return { type: "text", text: toString(contentItem.text) };
             }
+            if (contentItem.type === "refusal") {
+              return { type: "text", text: toString(contentItem.refusal) };
+            }
             if (contentItem.type === "input_image") {
               const imgResult: JsonRecord = {
                 type: "image_url",
@@ -330,6 +349,15 @@ export function openaiResponsesToOpenAIRequest(
       // Skip reasoning items - they are display-only metadata
       continue;
     }
+
+    if (itemType === "additional_tools") {
+      // Already consumed by collectResponsesTools() before message conversion.
+      continue;
+    }
+
+    throw unsupportedFeature(
+      `Unsupported Responses API feature: input item type '${itemType || "missing"}' cannot be represented in Chat Completions`
+    );
   }
 
   // Flush remainder
@@ -343,8 +371,8 @@ export function openaiResponsesToOpenAIRequest(
   }
 
   // Convert tools format
-  if (Array.isArray(root.tools)) {
-    result.tools = root.tools
+  if (tools.length > 0) {
+    result.tools = tools
       .filter((toolValue) => {
         const tool = toRecord(toolValue);
         const toolType = toString(tool.type);
@@ -519,7 +547,55 @@ export function openaiResponsesToOpenAIRequest(
       result.tool_choice = { type: "function", function: { name: tc.name } };
     } else if (tcType === "local_shell") {
       result.tool_choice = { type: "function", function: { name: "shell" } };
-    } else if (tcType && tcType !== "function" && tcType !== "allowed_tools") {
+    } else if (tcType === "allowed_tools") {
+      const mode = toString(tc.mode);
+      if (mode !== "auto" && mode !== "required") {
+        throw unsupportedFeature(
+          `Unsupported Responses API feature: allowed_tools mode '${mode || "missing"}' is not supported by omniroute`
+        );
+      }
+      if (!Array.isArray(tc.tools) || tc.tools.length === 0) {
+        throw unsupportedFeature(
+          "Unsupported Responses API feature: allowed_tools requires at least one function tool"
+        );
+      }
+
+      const allowedNames = new Set<string>();
+      for (const allowedValue of tc.tools) {
+        const allowed = toRecord(allowedValue);
+        const allowedType = toString(allowed.type);
+        const allowedName = toString(allowed.name).trim();
+        if (allowedType !== "function" || !allowedName) {
+          throw unsupportedFeature(
+            `Unsupported Responses API feature: allowed_tools descriptor type '${allowedType || "missing"}' cannot be represented in Chat Completions`
+          );
+        }
+        allowedNames.add(allowedName);
+      }
+
+      const chatTools = Array.isArray(result.tools) ? result.tools : [];
+      const availableNames = new Set(
+        chatTools
+          .map((toolValue) => toString(toRecord(toRecord(toolValue).function).name))
+          .filter(Boolean)
+      );
+      const missingNames = [...allowedNames].filter((name) => !availableNames.has(name));
+      if (missingNames.length > 0) {
+        throw unsupportedFeature(
+          `Unsupported Responses API feature: allowed_tools references unavailable function tool(s): ${missingNames.join(", ")}`
+        );
+      }
+
+      result.tools = chatTools.filter((toolValue) =>
+        allowedNames.has(toString(toRecord(toRecord(toolValue).function).name))
+      );
+      if (result.tools.length === 0) {
+        throw unsupportedFeature(
+          "Unsupported Responses API feature: allowed_tools resolved to zero Chat Completions function tools"
+        );
+      }
+      result.tool_choice = mode;
+    } else if (tcType && tcType !== "function") {
       // Built-in tool types (web_search_preview, file_search, etc.) have no Chat equivalent
       throw unsupportedFeature(
         `Unsupported Responses API feature: tool_choice type '${tcType}' is not supported by omniroute`
@@ -573,6 +649,12 @@ export function openaiResponsesToOpenAIRequest(
   // Completions equivalent. Strict non-OpenAI upstreams (e.g. NVIDIA NIM) reject
   // it with HTTP 400 "Unsupported parameter(s): truncation" (#2311).
   delete result.truncation;
+  // These fields configure Responses-owned state, caching, and tool execution limits.
+  // Chat Completions has no equivalent and strict compatible endpoints reject them.
+  delete result.max_tool_calls;
+  delete result.conversation;
+  delete result.prompt_cache_options;
+  delete result.prompt_cache_retention;
 
   return result;
 }
diff --git a/open-sse/translator/request/openai-responses/additionalTools.ts b/open-sse/translator/request/openai-responses/additionalTools.ts
new file mode 100644
index 0000000000..50cb27004f
--- /dev/null
+++ b/open-sse/translator/request/openai-responses/additionalTools.ts
@@ -0,0 +1,96 @@
+type JsonRecord = Record<string, unknown>;
+
+function toRecord(value: unknown): JsonRecord {
+  return value && typeof value === "object" && !Array.isArray(value) ? (value as JsonRecord) : {};
+}
+
+function toolName(value: unknown): string {
+  const tool = toRecord(value);
+  const nestedFunction = toRecord(tool.function);
+  return typeof tool.name === "string" && tool.name.trim()
+    ? tool.name.trim()
+    : typeof nestedFunction.name === "string" && nestedFunction.name.trim()
+      ? nestedFunction.name.trim()
+      : "";
+}
+
+function toolIdentity(value: unknown): string | null {
+  const tool = toRecord(value);
+  const name = toolName(value);
+  if (!name) return null;
+
+  // Namespace names are only containers. The Chat conversion flattens their nested
+  // tools, so they do not collide with a top-level function of the same name.
+  return tool.type === "namespace" ? `namespace:${name}` : `name:${name}`;
+}
+
+function mergeNamespaceTools(first: unknown, second: unknown): unknown[] {
+  const merged: unknown[] = [];
+  const seenNames = new Set<string>();
+
+  for (const source of [first, second]) {
+    if (!Array.isArray(source)) continue;
+    for (const tool of source) {
+      const name = toolName(tool);
+      if (name && seenNames.has(name)) continue;
+      if (name) seenNames.add(name);
+      merged.push(tool);
+    }
+  }
+
+  return merged;
+}
+
+/**
+ * Collect all Responses tool declarations before downgrading to Chat Completions.
+ *
+ * Most clients use the top-level `tools` array. Newer agent clients may instead add one or
+ * more `{ type: "additional_tools", tools: [...] }` input items so tool availability can be
+ * changed alongside the conversation transcript. Both forms describe tools available for
+ * the current response and therefore share the same downstream conversion path.
+ *
+ * Explicit top-level declarations take precedence on named collisions. Same-named namespaces are
+ * merged before deduplication because the Chat conversion flattens their members. Unnamed hosted
+ * tools are kept verbatim because their type can be repeated with distinct provider-specific
+ * configuration. This keeps the established request contract stable and prevents duplicate
+ * function names from reaching strict upstreams.
+ */
+export function collectResponsesTools(rootTools: unknown, inputItems: unknown[]): unknown[] {
+  const sources: unknown[][] = [Array.isArray(rootTools) ? rootTools : []];
+
+  for (const itemValue of inputItems) {
+    const item = toRecord(itemValue);
+    if (item.type === "additional_tools" && Array.isArray(item.tools)) {
+      sources.push(item.tools);
+    }
+  }
+
+  const merged: unknown[] = [];
+  const seen = new Set<string>();
+  const namespaceIndexes = new Map<string, number>();
+  for (const source of sources) {
+    for (const tool of source) {
+      const toolRecord = toRecord(tool);
+      if (toolRecord.type === "namespace") {
+        const namespaceName = toolName(tool);
+        const existingIndex = namespaceName ? namespaceIndexes.get(namespaceName) : undefined;
+        if (existingIndex !== undefined) {
+          const existing = toRecord(merged[existingIndex]);
+          merged[existingIndex] = {
+            ...toolRecord,
+            ...existing,
+            tools: mergeNamespaceTools(existing.tools, toolRecord.tools),
+          };
+          continue;
+        }
+        if (namespaceName) namespaceIndexes.set(namespaceName, merged.length);
+      }
+
+      const identity = toolIdentity(tool);
+      if (identity && seen.has(identity)) continue;
+      if (identity) seen.add(identity);
+      merged.push(tool);
+    }
+  }
+  return merged;
+}
diff --git a/tests/unit/responses-additional-tools.test.ts b/tests/unit/responses-additional-tools.test.ts
new file mode 100644
index 0000000000..517b1e94e0
--- /dev/null
+++ b/tests/unit/responses-additional-tools.test.ts
@@ -0,0 +1,223 @@
+import test from "node:test";
+import assert from "node:assert/strict";
+
+const { openaiResponsesToOpenAIRequest } =
+  await import("../../open-sse/translator/request/openai-responses.ts");
+
+interface ChatTool {
+  function: {
+    name: string;
+    description?: string;
+    parameters?: unknown;
+  };
+}
+
+interface ChatRequest {
+  messages: unknown[];
+  tools: ChatTool[];
+}
+
+test("Responses -> Chat merges additional_tools into the universal tool conversion path", () => {
+  const result = openaiResponsesToOpenAIRequest(
+    "any-model",
+    {
+      input: [
+        {
+          type: "additional_tools",
+          role: "developer",
+          tools: [
+            {
+              type: "custom",
+              name: "exec",
+              description: "Run orchestration code",
+              format: { type: "grammar", syntax: "lark", definition: "start: /.+/" },
+            },
+            {
+              type: "function",
+              name: "wait",
+              description: "Wait for a running operation",
+              parameters: {
+                type: "object",
+                properties: { cell_id: { type: "string" } },
+                required: ["cell_id"],
+              },
+            },
+            {
+              type: "namespace",
+              name: "collaboration",
+              tools: [
+                {
+                  name: "spawn_agent",
+                  description: "Spawn an agent",
+                  parameters: { type: "object", properties: {} },
+                },
+              ],
+            },
+          ],
+        },
+        {
+          type: "message",
+          role: "user",
+          content: [{ type: "input_text", text: "Use the tools" }],
+        },
+      ],
+      tool_choice: "auto",
+    },
+    true,
+    { provider: "any-openai-compatible-provider" }
+  ) as ChatRequest;
+
+  assert.deepEqual(result.messages, [
+    { role: "user", content: [{ type: "text", text: "Use the tools" }] },
+  ]);
+  assert.deepEqual(
+    result.tools.map((tool) => tool.function?.name),
+    ["exec", "wait", "spawn_agent"]
+  );
+  assert.deepEqual(result.tools[0].function.parameters, {
+    type: "object",
+    properties: { input: { type: "string" } },
+    required: ["input"],
+    additionalProperties: false,
+  });
+});
+
+test("Responses -> Chat merges multiple tool sources and keeps top-level declarations on conflict", () => {
+  const topLevel = {
+    type: "function",
+    name: "lookup",
+    description: "Authoritative top-level declaration",
+    parameters: { type: "object", properties: { id: { type: "string" } } },
+  };
+  const result = openaiResponsesToOpenAIRequest(
+    "any-model",
+    {
+      input: [
+        {
+          type: "additional_tools",
+          tools: [
+            { ...topLevel, description: "Conflicting deferred declaration" },
+            { type: "function", name: "first", parameters: { type: "object" } },
+          ],
+        },
+        {
+          type: "additional_tools",
+          tools: [{ type: "function", name: "second", parameters: { type: "object" } }],
+        },
+        { type: "message", role: "user", content: [{ type: "input_text", text: "go" }] },
+      ],
+      tools: [topLevel],
+    },
+    false,
+    { provider: "another-provider" }
+  ) as ChatRequest;
+
+  assert.deepEqual(
+    result.tools.map((tool) => tool.function.name),
+    ["lookup", "first", "second"]
+  );
+  assert.equal(result.tools[0].function.description, "Authoritative top-level declaration");
+});
+
+test("Responses -> Chat preserves a namespace that shares a name with a function", () => {
+  const result = openaiResponsesToOpenAIRequest(
+    "any-model",
+    {
+      input: [
+        {
+          type: "additional_tools",
+          tools: [
+            {
+              type: "namespace",
+              name: "server",
+              tools: [
+                {
+                  name: "mcp__server__read",
+                  parameters: { type: "object", properties: {} },
+                },
+              ],
+            },
+          ],
+        },
+        { type: "message", role: "user", content: [{ type: "input_text", text: "go" }] },
+      ],
+      tools: [{ type: "function", name: "server", parameters: { type: "object" } }],
+    },
+    false,
+    { provider: "another-provider" }
+  ) as ChatRequest;
+
+  assert.deepEqual(
+    result.tools.map((tool) => tool.function.name),
+    ["server", "mcp__server__read"]
+  );
+});
+
+test("Responses -> Chat merges members from same-named namespaces", () => {
+  const result = openaiResponsesToOpenAIRequest(
+    "any-model",
+    {
+      input: [
+        {
+          type: "additional_tools",
+          tools: [
+            {
+              type: "namespace",
+              name: "server",
+              tools: [
+                {
+                  name: "mcp__server__write",
+                  parameters: { type: "object", properties: {} },
+                },
+              ],
+            },
+          ],
+        },
+        { type: "message", role: "user", content: [{ type: "input_text", text: "go" }] },
+      ],
+      tools: [
+        {
+          type: "namespace",
+          name: "server",
+          tools: [
+            {
+              name: "mcp__server__read",
+              parameters: { type: "object", properties: {} },
+            },
+          ],
+        },
+      ],
+    },
+    false,
+    { provider: "another-provider" }
+  ) as ChatRequest;
+
+  assert.deepEqual(
+    result.tools.map((tool) => tool.function.name),
+    ["mcp__server__read", "mcp__server__write"]
+  );
+});
+
+test("Responses -> Chat validates tools supplied through additional_tools", () => {
+  assert.throws(
+    () =>
+      openaiResponsesToOpenAIRequest(
+        "any-model",
+        {
+          input: [
+            {
+              type: "additional_tools",
+              tools: [{ type: "file_search", name: "search" }],
+            },
+            { type: "message", role: "user", content: "hi" },
+          ],
+        },
+        false,
+        { provider: "any-provider" }
+      ),
+    (error: unknown) => {
+      const typedError = error as { statusCode?: number; errorType?: string };
+      return typedError.statusCode === 400 && typedError.errorType === "unsupported_feature";
+    }
+  );
+});
diff --git a/tests/unit/responses-chat-translation-gaps.test.ts b/tests/unit/responses-chat-translation-gaps.test.ts
new file mode 100644
index 0000000000..47ed3785d9
--- /dev/null
+++ b/tests/unit/responses-chat-translation-gaps.test.ts
@@ -0,0 +1,192 @@
+import test from "node:test";
+import assert from "node:assert/strict";
+
+const { openaiResponsesToOpenAIRequest } =
+  await import("../../open-sse/translator/request/openai-responses.ts");
+
+function translate(body: Record<string, unknown>): Record<string, unknown> {
+  return openaiResponsesToOpenAIRequest("gpt-5", body, false, null) as Record<string, unknown>;
+}
+
+test("Responses -> Chat preserves json_schema structured output", () => {
+  const result = translate({
+    input: "Return JSON",
+    text: {
+      format: {
+        type: "json_schema",
+        name: "answer",
+        description: "Structured answer",
+        schema: { type: "object", properties: { answer: { type: "string" } } },
+        strict: true,
+      },
+    },
+  });
+
+  assert.deepEqual(result.response_format, {
+    type: "json_schema",
+    json_schema: {
+      name: "answer",
+      description: "Structured answer",
+      schema: { type: "object", properties: { answer: { type: "string" } } },
+      strict: true,
+    },
+  });
+  assert.equal(result.text, undefined);
+});
+
+test("Responses -> Chat preserves json_object structured output", () => {
+  const result = translate({ input: "Return JSON", text: { format: { type: "json_object" } } });
+
+  assert.deepEqual(result.response_format, { type: "json_object" });
+  assert.equal(result.text, undefined);
+});
+
+test("Responses -> Chat restricts tools selected by allowed_tools", () => {
+  const result = translate({
+    input: "Use one tool",
+    tools: [
+      { type: "function", name: "keep", parameters: { type: "object" } },
+      { type: "function", name: "remove", parameters: { type: "object" } },
+    ],
+    tool_choice: {
+      type: "allowed_tools",
+      mode: "required",
+      tools: [{ type: "function", name: "keep" }],
+    },
+  });
+
+  assert.equal(result.tool_choice, "required");
+  assert.deepEqual(
+    (result.tools as Array<{ function: { name: string } }>).map((tool) => tool.function.name),
+    ["keep"]
+  );
+});
+
+test("Responses -> Chat resolves allowed_tools against flattened namespace tools", () => {
+  const result = translate({
+    input: "Use a namespaced tool",
+    tools: [
+      {
+        type: "namespace",
+        name: "server",
+        tools: [{ name: "mcp__server__read", parameters: { type: "object" } }],
+      },
+    ],
+    tool_choice: {
+      type: "allowed_tools",
+      mode: "auto",
+      tools: [{ type: "function", name: "mcp__server__read" }],
+    },
+  });
+
+  assert.equal(result.tool_choice, "auto");
+  assert.equal(
+    (result.tools as Array<{ function: { name: string } }>)[0].function.name,
+    "mcp__server__read"
+  );
+});
+
+test("Responses -> Chat rejects malformed or unavailable allowed_tools", () => {
+  assert.throws(
+    () =>
+      translate({
+        input: "Use a tool",
+        tools: [{ type: "function", name: "available", parameters: { type: "object" } }],
+        tool_choice: {
+          type: "allowed_tools",
+          mode: "required",
+          tools: [{ type: "function", name: "missing" }],
+        },
+      }),
+    (error: unknown) =>
+      error instanceof Error &&
+      (error as Error & { errorType?: string }).errorType === "unsupported_feature"
+  );
+
+  assert.throws(
+    () =>
+      translate({
+        input: "Use a tool",
+        tools: [{ type: "function", name: "available", parameters: { type: "object" } }],
+        tool_choice: {
+          type: "allowed_tools",
+          mode: "required",
+          tools: [{ type: "web_search", name: "available" }],
+        },
+      }),
+    (error: unknown) =>
+      error instanceof Error &&
+      (error as Error & { errorType?: string }).errorType === "unsupported_feature"
+  );
+});
+
+test("Responses -> Chat rejects input item types without a lossless Chat equivalent", () => {
+  for (const item of [
+    { type: "item_reference", id: "item_123" },
+    { type: "computer_call_output", call_id: "call_1", output: {} },
+    { type: "mcp_call", name: "remote", arguments: "{}" },
+    { type: "web_search_call", id: "search_1" },
+    { unexpected: true },
+  ]) {
+    assert.throws(
+      () => translate({ input: [item] }),
+      (error: unknown) =>
+        error instanceof Error &&
+        (error as Error & { errorType?: string }).errorType === "unsupported_feature" &&
+        error.message.includes("input item type")
+    );
+  }
+});
+
+test("Responses -> Chat consumes additional_tools input items without emitting messages", () => {
+  const result = translate({
+    input: [
+      { type: "message", role: "user", content: [{ type: "input_text", text: "Use it" }] },
+      {
+        type: "additional_tools",
+        tools: [{ type: "function", name: "extra", parameters: { type: "object" } }],
+      },
+    ],
+  });
+
+  assert.equal((result.messages as unknown[]).length, 1);
+  assert.equal((result.tools as Array<{ function: { name: string } }>)[0].function.name, "extra");
+});
+
+test("Responses -> Chat converts refusal history to valid Chat text content", () => {
+  const result = translate({
+    input: [
+      {
+        type: "message",
+        role: "assistant",
+        content: [{ type: "refusal", refusal: "I cannot help with that." }],
+      },
+    ],
+  });
+
+  assert.deepEqual(result.messages, [
+    {
+      role: "assistant",
+      content: [{ type: "text", text: "I cannot help with that." }],
+    },
+  ]);
+});
+
+test("Responses -> Chat strips Responses-only execution and cache fields", () => {
+  const result = translate({
+    input: "Hello",
+    max_tool_calls: 3,
+    conversation: "conv_123",
+    prompt_cache_options: { retention: "24h" },
+    prompt_cache_retention: "24h",
+    metadata: { keep: true },
+    parallel_tool_calls: true,
+  });
+
+  assert.equal(result.max_tool_calls, undefined);
+  assert.equal(result.conversation, undefined);
+  assert.equal(result.prompt_cache_options, undefined);
+  assert.equal(result.prompt_cache_retention, undefined);
+  assert.deepEqual(result.metadata, { keep: true });
+  assert.equal(result.parallel_tool_calls, true);
+});