diff --git a/backend/aci/server/agent/prompt.py b/backend/aci/server/agent/prompt.py
index 6d9e68c64..cafa155ae 100644
--- a/backend/aci/server/agent/prompt.py
+++ b/backend/aci/server/agent/prompt.py
@@ -12,6 +12,11 @@
 
 logger = get_logger(__name__)
 
+# OpenAI API has limits on tool count and total token size of tool definitions.
+# Empirically, requests fail around 128+ tools or when tool definitions exceed ~200k tokens.
+# Using a conservative limit to prevent cryptic API errors.
+MAX_TOOLS = 64
+
 
 def convert_to_openai_messages(messages: list[ClientMessage]) -> list[ChatCompletionMessageParam]:
     """
@@ -69,7 +74,19 @@ async def openai_chat_stream(
     Args:
         messages: List of chat messages
         tools: List of tools to use
+
+    Raises:
+        ValueError: If too many tools are selected, exceeding API limits.
     """
+    if len(tools) > MAX_TOOLS:
+        logger.warning(
+            f"Tool limit exceeded: {len(tools)} tools requested, max is {MAX_TOOLS}"
+        )
+        raise ValueError(
+            f"Too many tools selected ({len(tools)}). Maximum allowed is {MAX_TOOLS}. "
+            "Please reduce the number of selected functions or apps."
+        )
+
     client = OpenAI(api_key=config.OPENAI_API_KEY)
 
     # TODO: support different meta function mode ACI_META_FUNCTIONS_SCHEMA_LIST
diff --git a/backend/aci/server/routes/agent.py b/backend/aci/server/routes/agent.py
index c3adf9541..509f3ab65 100644
--- a/backend/aci/server/routes/agent.py
+++ b/backend/aci/server/routes/agent.py
@@ -1,6 +1,6 @@
 from typing import Annotated
 
-from fastapi import APIRouter, Depends
+from fastapi import APIRouter, Depends, HTTPException
 from fastapi.responses import StreamingResponse
 from openai import OpenAI
 from pydantic import BaseModel
@@ -12,6 +12,7 @@
 from aci.server import dependencies as deps
 from aci.server.agent.prompt import (
     ClientMessage,
+    MAX_TOOLS,
     convert_to_openai_messages,
     openai_chat_stream,
 )
@@ -62,6 +63,13 @@ async def handle_chat(
         func for func in selected_functions if isinstance(func, OpenAIResponsesFunctionDefinition)
     ]
 
+    if len(tools) > MAX_TOOLS:
+        raise HTTPException(
+            status_code=400,
+            detail=f"Too many tools selected ({len(tools)}). Maximum allowed is {MAX_TOOLS}. "
+            "Please reduce the number of selected functions or apps.",
+        )
+
     response = StreamingResponse(openai_chat_stream(openai_messages, tools=tools))
     response.headers["x-vercel-ai-data-stream"] = "v1"