Why Agent

Tue, 30 Sep 2025 11:54:06 +0800

I’ve always had a question: Why do we need agent frameworks? Aren’t large models enough on their own? This article reflects my current understanding of the subject.

After using several tools extensively and participating in multiple agent projects recently, I’ve reached some conclusions.

The Limitations of LLMs

The primary reason for using agents is the inherent limitations of LLMs.

First and foremost is the context window, as explicitly mentioned in langchain/subagent. Although many modern models have significantly expanded context windows (GPT-4 Turbo 128K, Claude-3.5 Sonnet 200K, Gemini-1.5 Pro up to 2M), they are still insufficient for truly complex tasks. For example, processing a massive codebase or analyzing hundreds of documents quickly exhausts these limits. Furthermore, processing extremely long contexts is both expensive and slow.

Beyond context, there are other capability gaps:

Vision Capabilities: While modern VLMs (Vision Language Models) are powerful, traditional CV (Computer Vision) models often perform better in specific scenarios. Additionally, some models (like DeepSeek-V3) don’t have native vision capabilities.
Resource Access: LLMs cannot directly interact with databases, file systems, or network services.
Specialized Tools: Tools for code execution, complex mathematics, or data analysis require protocols like MCP to be accessible to an LLM.

What Agents Can Do

Beyond addressing the limitations above, here are some practical ways agents add value.

Domain-Specific Text Processing

Agents can process different text segments (contexts) independently.

Context Optimization: Agents can compress or selectively provide context, effectively extending the usable context window.
Performance Gains: An LLM within an agent can focus on a single, specific task, leading to better performance. When given too much text, LLMs often struggle to identify key information; smaller, targeted context makes this much easier.
Specialized Knowledge: LLMs are trained on general data. To make an agent a domain expert, we can inject specific knowledge directly into its context.

Visual Capability Integration

Through agents, we can integrate traditional vision models to handle tasks that LLMs struggle with. For example, using an MCP (Model Context Protocol) to bridge an agent with vision capabilities.

A notable example is Zhipu’s Vision MCP. Using this MCP in conjunction with an agent significantly enhances visual processing power. This highlights the value of MCP servers that integrate specialized services.

Agent Frameworks

Pydantic AI: I find this particularly useful because it integrates Pydantic models into the agent framework, making it much easier to debug. I’ve tested its integration with Qwen3.
LangChain: I haven’t used this in production, only for basic debugging. The API changes frequently, which can be challenging. One minor issue is prompt handling; I used Jinja to solve this. Alternatively, the “LangChain way” involves using PromptTemplates.

Work With Langfuse

Mon, 21 Apr 2025 14:51:38 +0800

When developing LLM applications, we consider performance issues during LLM calls and monitor outputs during the process.

At this point, tools like LangSmith and Langfuse become very useful.

However, sometimes we have local computing resources and prefer not to use cloud-based resources for LLM call monitoring, so we might not consider LangSmith.

In such cases, we can use Langfuse for this purpose.

Deployment

Deploying Langfuse is very simple; all you need to do is:

1
2
3


git clone https://github.com/langfuse/langfuse.git
cd langfuse
docker compose up -d

This way, the deployment is successful.

Replacement

If you previously used OpenAI’s SDK, you can continue using it as follows.

Install langfuse in the project:

1

pip install langfuse

To configure the API key, you need to use it in the deployed langfuse:

1
2
3


LANGFUSE_SECRET_KEY=<secret key>
LANGFUSE_PUBLIC_KEY=<public key>
LANGFUSE_HOST="http://localhost:3001"

Here I have set the Langfuse port to 3001; you should adjust according to your own configuration.

Simply replace the original OpenAI configuration:

1
2
3


# remove: import openai

from langfuse.openai import openai

In addition, langfuse also supports langchain and llamaindex, which will not be elaborated on further here.

Thoughts

Coze is also developing a large model agent framework, but the approach is quite different. Coze is building everything, including workflows and LLMs, making it relatively closed.

However, langfuse is more open, allowing the use of langchain and other models.

As a developer from a small company, I prefer the langfuse model because it offers more choices. However, if the project timeline is tight and Coze is barely usable, I would choose Coze.

Issues

An exception occurred when I replaced the OpenAI SDK:

1

Unexpected error occurred. Please check your request and contact support: https://langfuse.com/support.

I still encountered issues when testing test_langfuse.py:

 1
 2
 3
 4
 5
 6
 7
 8
 9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31


import os

from langfuse.decorators import observe
from langfuse.openai import openai


@observe()
def story():
 return (
 openai.chat.completions.create(
 model="moonshot-v1-auto",
 max_tokens=100,
 messages=[
 {"role": "system", "content": "You are a great storyteller."},
 {"role": "user", "content": "Once upon a time in a galaxy far, far away..."},
 ],
 )
 .choices[0]
 .message.content
 )


@observe()
def main():
 return story()


def test_langfuse():
 assert os.getenv("OPENAI_BASE_URL") is not None
 assert os.getenv("OPENAI_API_KEY") is not None
 main()

Regarding this issue, I have opened a discussion.

Additionally, if you wish to view the original code, you can obtain it from https://github.com/svtter/pdf-reader.

Agent on Svtter's Blog