diff --git a/README.md b/README.md
index 06b90381a..4ff9e72db 100644
--- a/README.md
+++ b/README.md
@@ -128,7 +128,7 @@ mkdir acontext_server && cd acontext_server
 acontext server up
 ```
 
-> Make sure your LLM has the ability to [call tools](https://platform.openai.com/docs/guides/function-calling). By default, Acontext will use `gpt-4.1`.
+> Make sure your LLM has the ability to [call tools](https://platform.openai.com/docs/guides/function-calling). By default, Acontext will use `gpt-4.1`. You can also use other LLM providers by setting `LLM_SDK` to `anthropic` in your `.env` file, or use any OpenAI-compatible provider (e.g., MiniMax, DeepSeek) by setting `LLM_BASE_URL`.
 
 `acontext server up` will create/use `.env` and `config.yaml` for Acontext, and create a `db` folder to persist data.
 
diff --git a/docs/content/docs/(guides)/settings/core.mdx b/docs/content/docs/(guides)/settings/core.mdx
index d5f37f3b5..753403817 100644
--- a/docs/content/docs/(guides)/settings/core.mdx
+++ b/docs/content/docs/(guides)/settings/core.mdx
@@ -29,6 +29,10 @@ Default model identifier for LLM operations. Examples: `gpt-4`, `gpt-3.5-turbo`,
 Timeout in seconds for LLM API responses. Increase for longer operations.
 </ParamField>
 
+<ParamField path="LLM_STRIP_TAGS" type="string" default="">
+Comma-separated list of XML-style tag names to strip from LLM responses. Many reasoning models wrap chain-of-thought in tags like `<think>...</think>`. Set to `think` to strip those blocks, or `think,reasoning` for multiple tag types. Default is empty (no stripping — original content preserved).
+</ParamField>
+
 ### Embedding Configuration
 
 <ParamField path="BLOCK_EMBEDDING_PROVIDER" type="string" default="openai">
@@ -89,6 +93,21 @@ BLOCK_EMBEDDING_PROVIDER=openai
 BLOCK_EMBEDDING_API_KEY=sk-your-openai-key-for-embeddings
 ```
 
+```bash title="MiniMax Setup"
+# MiniMax is OpenAI-compatible — use LLM_SDK=openai with a custom base URL
+LLM_API_KEY=your-minimax-api-key
+LLM_SDK=openai
+LLM_BASE_URL=https://api.minimax.io/v1
+LLM_SIMPLE_MODEL=MiniMax-M2.7
+
+# Strip reasoning tags from MiniMax responses (optional)
+LLM_STRIP_TAGS=think
+
+# Keep OpenAI for embeddings (recommended)
+BLOCK_EMBEDDING_PROVIDER=openai
+BLOCK_EMBEDDING_API_KEY=sk-your-openai-key-for-embeddings
+```
+
 ```bash title="Custom Endpoints"
 # Custom LLM endpoint (e.g., Azure OpenAI)
 LLM_API_KEY=your-azure-key
diff --git a/src/server/core/acontext_core/llm/complete/openai_sdk.py b/src/server/core/acontext_core/llm/complete/openai_sdk.py
index fdefff8aa..6a4bef7c7 100644
--- a/src/server/core/acontext_core/llm/complete/openai_sdk.py
+++ b/src/server/core/acontext_core/llm/complete/openai_sdk.py
@@ -1,4 +1,5 @@
 import json
+import re
 from typing import Optional
 from .clients import get_openai_async_client_instance
 from openai.types.chat import ChatCompletion
@@ -9,6 +10,23 @@
 from ...telemetry.log import get_wide_event
 
 
+def _strip_tags(text: str, tags: list[str]) -> str:
+    """Strip named XML-style tag blocks from model responses.
+
+    Many reasoning models (DeepSeek, QwQ, MiniMax, etc.) wrap their internal
+    chain-of-thought in tags like ``<think>...</think>``.  This helper removes
+    the specified tag blocks so that downstream consumers receive only the
+    final answer.
+
+    Args:
+        text: The raw response text.
+        tags: Tag names to strip, e.g. ``["think", "reasoning"]``.
+    """
+    for tag in tags:
+        text = re.sub(rf"<{tag}>[\s\S]*?</{tag}>\s*", "", text)
+    return text.strip()
+
+
 def convert_openai_tool_to_llm_tool(tool_body: ChatCompletionMessageToolCall) -> dict:
     return {
         "id": tool_body.id,
@@ -90,20 +108,24 @@ async def openai_complete(
         else None
     )
 
+    content = response.choices[0].message.content
+    if content and DEFAULT_CORE_CONFIG.llm_strip_tags:
+        content = _strip_tags(content, DEFAULT_CORE_CONFIG.llm_strip_tags)
+
     llm_response = LLMResponse(
         role=response.choices[0].message.role,
         raw_response=response,
-        content=response.choices[0].message.content,
+        content=content,
         tool_calls=_tu,
     )
 
     if json_mode:
         try:
-            json_content = json.loads(response.choices[0].message.content)
+            json_content = json.loads(content) if content else None
         except json.JSONDecodeError:
             LOG.error(
                 "llm.json_decode_error",
-                content=response.choices[0].message.content[:200],
+                content=(content or "")[:200],
             )
             json_content = None
         llm_response.json_content = json_content
diff --git a/src/server/core/acontext_core/schema/config.py b/src/server/core/acontext_core/schema/config.py
index f5ec77287..51ee2ec6a 100644
--- a/src/server/core/acontext_core/schema/config.py
+++ b/src/server/core/acontext_core/schema/config.py
@@ -1,6 +1,6 @@
 import os
 import yaml
-from pydantic import BaseModel
+from pydantic import BaseModel, field_validator
 from typing import Literal, Mapping, Optional, Any, Type
 
 
@@ -24,6 +24,15 @@ class CoreConfig(BaseModel):
     llm_sdk: Literal["openai", "anthropic", "mock"] = "openai"
 
     llm_simple_model: str = "gpt-4.1"
+    llm_strip_tags: list[str] = []
+
+    @field_validator("llm_strip_tags", mode="before")
+    @classmethod
+    def parse_strip_tags(cls, v):
+        """Accept a comma-separated string (from env var) or a list."""
+        if isinstance(v, str):
+            return [t.strip() for t in v.split(",") if t.strip()]
+        return v
 
     # Core Configuration
     logging_format: str = "json"
diff --git a/src/server/core/tests/llm/test_think_tag_stripping.py b/src/server/core/tests/llm/test_think_tag_stripping.py
new file mode 100644
index 000000000..0eae0f241
--- /dev/null
+++ b/src/server/core/tests/llm/test_think_tag_stripping.py
@@ -0,0 +1,219 @@
+"""
+Tests for configurable tag stripping in ``openai_complete``.
+
+The ``_strip_tags`` helper removes XML-style tag blocks (e.g.
+``<think>...</think>``, ``<reasoning>...</reasoning>``) that reasoning models
+inject before their final answer.  Stripping is **off by default** and
+controlled by the ``llm_strip_tags`` config field (env var
+``LLM_STRIP_TAGS``).
+"""
+
+import json
+import pytest
+from unittest.mock import AsyncMock, patch, MagicMock
+from openai.types.chat import ChatCompletion, ChatCompletionMessage
+from openai.types.chat.chat_completion import Choice
+from openai.types.completion_usage import CompletionUsage
+
+from acontext_core.llm.complete.openai_sdk import _strip_tags, openai_complete
+from acontext_core.schema.llm import LLMResponse
+
+
+# ---------------------------------------------------------------------------
+# Helpers
+# ---------------------------------------------------------------------------
+
+def _make_chat_completion(content="Hello", tool_calls=None):
+    """Build a real ``ChatCompletion`` object for tests."""
+    message = ChatCompletionMessage(
+        role="assistant",
+        content=content,
+        tool_calls=tool_calls,
+    )
+    return ChatCompletion(
+        id="chatcmpl-test",
+        choices=[Choice(finish_reason="stop", index=0, message=message)],
+        created=1700000000,
+        model="test-model",
+        object="chat.completion",
+        usage=CompletionUsage(
+            prompt_tokens=10,
+            completion_tokens=20,
+            total_tokens=30,
+        ),
+    )
+
+
+def _patch_openai_complete(mock_response, strip_tags=None):
+    """Return context-manager patches for ``openai_complete``.
+
+    Args:
+        mock_response: The ``ChatCompletion`` to return from the mocked client.
+        strip_tags: List of tag names for ``llm_strip_tags`` config.
+                    Defaults to ``[]`` (no stripping).
+    """
+    if strip_tags is None:
+        strip_tags = []
+
+    mock_client = AsyncMock()
+    mock_client.chat.completions.create = AsyncMock(return_value=mock_response)
+
+    mock_cfg = MagicMock()
+    mock_cfg.llm_strip_tags = strip_tags
+    mock_cfg.llm_response_timeout = 60
+    mock_cfg.llm_openai_completion_kwargs = {}
+
+    p_client = patch(
+        "acontext_core.llm.complete.openai_sdk.get_openai_async_client_instance",
+        return_value=mock_client,
+    )
+    p_wide = patch(
+        "acontext_core.llm.complete.openai_sdk.get_wide_event",
+        return_value={},
+    )
+    p_config = patch(
+        "acontext_core.llm.complete.openai_sdk.DEFAULT_CORE_CONFIG",
+        mock_cfg,
+    )
+    return p_client, p_wide, p_config
+
+
+# ---------------------------------------------------------------------------
+# _strip_tags unit tests
+# ---------------------------------------------------------------------------
+
+class TestStripTags:
+    """Test stripping arbitrary XML-style tag blocks from model responses."""
+
+    def test_strip_single_tag(self):
+        text = "<think>reasoning here</think>actual response"
+        assert _strip_tags(text, ["think"]) == "actual response"
+
+    def test_strip_multiline_tag(self):
+        text = "<think>\nstep 1\nstep 2\n</think>\nfinal answer"
+        assert _strip_tags(text, ["think"]) == "final answer"
+
+    def test_no_matching_tags(self):
+        text = "just a normal response"
+        assert _strip_tags(text, ["think"]) == "just a normal response"
+
+    def test_empty_tag_block(self):
+        text = "<think></think>response"
+        assert _strip_tags(text, ["think"]) == "response"
+
+    def test_tag_in_middle(self):
+        text = "before <think>thinking</think> after"
+        assert _strip_tags(text, ["think"]) == "before  after"
+
+    def test_multiple_occurrences(self):
+        text = "<think>first</think>middle<think>second</think>end"
+        assert _strip_tags(text, ["think"]) == "middleend"
+
+    def test_nested_angle_brackets_inside_tag(self):
+        text = "<think>if a < b and b > c then</think>answer"
+        assert _strip_tags(text, ["think"]) == "answer"
+
+    def test_empty_string(self):
+        assert _strip_tags("", ["think"]) == ""
+
+    def test_only_tag_block(self):
+        text = "<think>all reasoning</think>"
+        assert _strip_tags(text, ["think"]) == ""
+
+    def test_multiple_tag_types(self):
+        text = "<think>thought</think>middle<reasoning>reason</reasoning>end"
+        assert _strip_tags(text, ["think", "reasoning"]) == "middleend"
+
+    def test_empty_tags_list_preserves_content(self):
+        text = "<think>reasoning</think>answer"
+        assert _strip_tags(text, []) == "<think>reasoning</think>answer"
+
+    def test_non_matching_tag_preserved(self):
+        text = "<think>reasoning</think>answer"
+        assert _strip_tags(text, ["reasoning"]) == "<think>reasoning</think>answer"
+
+
+# ---------------------------------------------------------------------------
+# openai_complete integration tests (mocked client)
+# ---------------------------------------------------------------------------
+
+class TestOpenAICompleteTagStripping:
+    """Verify that ``openai_complete`` strips tags only when configured."""
+
+    @pytest.mark.asyncio
+    async def test_no_stripping_by_default(self):
+        """With default config (empty strip_tags), think tags are preserved."""
+        raw = "<think>Let me reason...</think>The answer is 42."
+        response = _make_chat_completion(content=raw)
+        p1, p2, p3 = _patch_openai_complete(response, strip_tags=[])
+
+        with p1, p2, p3:
+            result = await openai_complete(prompt="test", model="m")
+
+        assert result.content == raw
+
+    @pytest.mark.asyncio
+    async def test_stripping_when_configured(self):
+        """When llm_strip_tags=["think"], think tags are removed."""
+        response = _make_chat_completion(
+            content="<think>Let me reason step by step...</think>The answer is 42."
+        )
+        p1, p2, p3 = _patch_openai_complete(response, strip_tags=["think"])
+
+        with p1, p2, p3:
+            result = await openai_complete(prompt="test", model="m")
+
+        assert isinstance(result, LLMResponse)
+        assert result.content == "The answer is 42."
+
+    @pytest.mark.asyncio
+    async def test_no_think_tags_unchanged(self):
+        """Response without think tags should pass through unchanged."""
+        response = _make_chat_completion(content="Hello from the model!")
+        p1, p2, p3 = _patch_openai_complete(response, strip_tags=["think"])
+
+        with p1, p2, p3:
+            result = await openai_complete(prompt="Say hello", model="m")
+
+        assert result.content == "Hello from the model!"
+
+    @pytest.mark.asyncio
+    async def test_json_mode_with_stripping(self):
+        """JSON mode should parse correctly after stripping think tags."""
+        response = _make_chat_completion(
+            content='<think>reasoning</think>{"key": "value"}'
+        )
+        p1, p2, p3 = _patch_openai_complete(response, strip_tags=["think"])
+
+        with p1, p2, p3:
+            result = await openai_complete(
+                prompt="Return JSON", model="m", json_mode=True,
+            )
+
+        assert result.json_content == {"key": "value"}
+
+    @pytest.mark.asyncio
+    async def test_none_content_not_stripped(self):
+        """None content (e.g. tool-call-only response) should remain None."""
+        response = _make_chat_completion(content=None)
+        p1, p2, p3 = _patch_openai_complete(response, strip_tags=["think"])
+
+        with p1, p2, p3:
+            result = await openai_complete(prompt="call a tool", model="m")
+
+        assert result.content is None
+
+    @pytest.mark.asyncio
+    async def test_multiple_tag_types_stripped(self):
+        """Multiple tag types configured should all be stripped."""
+        response = _make_chat_completion(
+            content="<think>thought</think><reasoning>reason</reasoning>final"
+        )
+        p1, p2, p3 = _patch_openai_complete(
+            response, strip_tags=["think", "reasoning"]
+        )
+
+        with p1, p2, p3:
+            result = await openai_complete(prompt="test", model="m")
+
+        assert result.content == "final"