Metadata-Version: 2.4
Name: langlens
Version: 0.1.0
Summary: A visual tracer and debugger for LangChain and LangGraph
Requires-Python: >=3.10
Description-Content-Type: text/markdown
License-File: LICENSE
Requires-Dist: langchain-core
Provides-Extra: demo
Requires-Dist: langchain[google-genai]>=1.2.2; extra == "demo"
Requires-Dist: langchain-ollama>=1.0.1; extra == "demo"
Requires-Dist: langgraph-cli[inmem]>=0.4.11; extra == "demo"
Requires-Dist: langchain-groq>=1.1.1; extra == "demo"
Requires-Dist: python-dotenv; extra == "demo"
Dynamic: license-file

<p align="center">
  <img src="web-ui/static/logo.svg" alt="LangLens Logo" width="400">
</p>

# LangLens

**Node-Level Observability for LangChain and LangGraph.**

LangLens is a lightweight visual tracer and debugger designed specifically for LLM-powered applications. It captures detailed execution logs from your agents and chains, allowing you to inspect every LLM call, tool execution, and state transition with precision.

![LangLens Screenshot](assests/screenshoot.png)

---

## ✨ Features

- 🔍 **Interactive Trace Viewer**: Explore complex workflows through a rich graphical interface.
- 🧩 **Node-Level Granularity**: Deep dive into individual chain runs, LLM inputs/outputs, and tool executions.
- 💻 **VS Code Extension**: Debug your traces directly inside VS Code with a custom editor experience.
- 🚀 **CLI Tool**: Quickly host a local viewer for any `.langlens` file.
- 🔌 **Seamless Integration**: Add observability to your existing projects with a single line of code.

---

## 🏗️ How it Works

LangLens bridges the gap between your LLM execution and visual debugging:

```mermaid
graph TD
    subgraph "Capture Phase"
    A[LLM App] -- "1. LangChain Callbacks" --> B(LangLens Handler)
    B -- "2. Stream Events" --> C[[.langlens file]]
    end

    subgraph "Visualization Phase"
    C -- "3a. CLI Server" --> D[Web Browser]
    C -- "3b. Custom Editor" --> E[VS Code IDE]

    D --> F{Interactive UI}
    E --> F

    F -- "4. Inspect Nodes" --> G[Observability & Debugging]
    end
```

1. **Capture**: The `LangLensCallbackHandler` hooks into LangChain's event system to record starts, ends, and errors of every chain, tool, and LLM call.
2. **Storage**: Events are serialized and saved into a `.langlens` file (JSONL format), providing a portable trace of the entire execution.
3. **Visualization**:
   - The **CLI Tool** launches a local Python server that hosts a Svelte-based frontend.
   - The **VS Code Extension** registers as a custom editor for `.langlens` files, embedding the same Svelte UI directly into your workspace.
4. **Debugging**: Use the interactive graph to navigate complex agentic workflows and inspect raw inputs/outputs at any level of the tree.

---

## 📦 Installation

Since LangLens is currently in development and not yet published to PyPI or the VS Code Marketplace, you can install it from source.

### 1. Build the Web UI

The Python viewer and VS Code extension both require the bundled Web UI.

```bash
# Build the UI and copy assets to the Python/VS Code directories
./build_ui.sh
```

### 2. Python Package

Install the package in editable mode from the root directory:

```bash
pip install -e .
```

### 3. VS Code Extension

To use the extension locally:

1. Open the `/vscode-extension` folder in VS Code.
2. Run `pnpm install`.
3. Press `F5` to start a new Extension Development Host window with the extension loaded.

## 📦 Packaging and Releases

To build both the Python package and the VS Code extension for distribution, you can use the provided script:

```bash
./scripts/package.sh
```

This will:

1. Build the Web UI.
2. Generate Python `.whl` and `.tar.gz` files in `dist/`.
3. Generate a VS Code `.vsix` file in `dist/`.

### Continuous Integration

This project includes a GitHub Action workflow in `.github/workflows/release.yml`. When you push a tag (e.g., `v0.1.0`), it will automatically:

- Build all packages.
- Create a GitHub Release with the artifacts attached.

## 🚀 Quick Start

### 1. Integrate the Callback

Add the `LangLensCallbackHandler` to your LangChain or LangGraph execution config.

```python
from langlens import LangLensCallbackHandler
from langchain_openai import ChatOpenAI

# Initialize the handler (creates a logs.langlens file)
handler = LangLensCallbackHandler(filename="logs.langlens")

# Attach it to your run
llm = ChatOpenAI()
config = {"callbacks": [handler]}
response = llm.invoke("How does a sunrise look from Mars?", config=config)
```

### 2. Visualize the Trace

#### Option A: Using the CLI

Run the following command to start a local server and open the viewer in your browser:

```bash
langlens visualize logs.langlens
```

#### Option B: Using VS Code

Simply open any `.langlens` file. The **LangLens Viewer** will automatically render the interactive trace.

## 🛠️ Development

### Building the Web UI

The UI is built with Svelte and Tailwind CSS.

```bash
cd web-ui
pnpm install
pnpm build
```

### Building the VS Code Extension

```bash
cd vscode-extension
pnpm install
pnpm run package
```

---

## 📄 License

This project is licensed under the MIT License.
