Open Interpreter is an open-source tool that allows developers to work with large language models (LLMs) on their local machine. It can be used via a terminal interface and supports popular programming languages, including Python, JavaScript, and Bash.
With Open Interpreter, users can perform various tasks, such as conducting data analysis, controlling web browsers for research purposes, generating and editing photos, videos, and PDFs, as well as managing and analyzing large datasets.
One of the most important features of Open Interpreter is its ability to control a PC's graphical user interface (GUI), allowing for direct interaction with the desktop environment. Additionally, it now includes vision capabilities, enabling it to analyze and interpret images.
Open Interpreter is open sourced under the AGPL-3.0 license. It has over 50K stars on GitHub and has over 100 contributors. Get Open Interpreter from the official GitHub repo
This is part of a series of articles about code interpreter (content to come).
In this article:
OpenAI’s Code Interpreter is a tool integrated with GPT-3.5 and GPT-4, which can perform various computational tasks using natural language. However, it operates in a hosted, closed-source environment with several restrictions, such as limited access to pre-installed packages and the absence of internet connectivity. Additionally, it has a runtime limit of 120 seconds and a file upload cap of 100MB, which can be limiting for extensive or complex tasks.
Open Interpreter offers a more flexible solution by running directly on your local machine. This local execution provides full access to the internet and allows the use of any package or library needed for your projects. With no restrictions on runtime or file size, Open Interpreter is well-suited for handling large datasets and lengthy computations. Its open-source nature ensures that you have complete control over the tool and your data. It also supports multiple large language models (LLMs) beyond those offered by OpenAI.
Open Interpreter uses LiteLLM to connect the terminal interface or development environment to hosted language models. When running open interpreter, users can determine what LLM model to run, like this:
interpreter --model gpt-4-turbo interpreter --model claude-3
You can see a full list of supported LLMs here. The following LLM providers are supported:
| Provider | Special Comments |
|---|---|
| OpenAI | Supports Chat + Embedding calls |
| OpenAI (Text Completion) | Supports text completion models |
| OpenAI-Compatible Endpoints | Allows models hosted behind an OpenAI proxy |
| Azure OpenAI | API Keys, Params |
| Azure AI Studio | Supports all models on Azure AI Studio |
| VertexAI | Supports Anthropic, Gemini, Model Garden |
| PaLM API - Google | |
| Gemini - Google AI Studio | |
| Anthropic | Supports all Anthropic models |
| AWS Sagemaker | Supports all Sagemaker Huggingface Jumpstart Models |
| AWS Bedrock | Supports Anthropic, Amazon Titan, A121 LLMs |
| Mistral AI API | |
| Codestral API | Available in select code-completion plugins, direct query support |
| Cohere | API KEYS |
| Anyscale | |
| Huggingface | Supports various Huggingface models |
| Databricks | Supports all models on Databricks |
| IBM watsonx.ai | Supports all IBM watsonx.ai foundational models and embeddings |
| Predibase | Supports all models on Predibase |
| Nvidia NIM | |
| Volcano Engine (Volcengine) | |
| Triton Inference Server | Supports Embedding Models on Triton Inference Servers |
| Ollama | Supports all models from Ollama |
| Perplexity AI (pplx-api) | |
| Groq | |
| Deepseek | |
| Fireworks AI | |
| Clarifai | Supports Anthropic, OpenAI, Mistral, Llama, and Gemini LLMs |
| VLLM | Supports all models on VLLM |
| Xinference (Xorbits Inference) | |
| Cloudflare Workers AI | |
| DeepInfra | |
| AI21 | Supports j2-light, j2-mid, and j2-ultra from AI21 |
| NLP Cloud | Supports all LLMs on NLP Cloud |
| Replicate | Supports all models on Replicate |
| Together AI | Supports all models on Together AI |
| Voyage AI | |
| Aleph Alpha | Supports all models from Aleph Alpha |
| Baseten | Supports any Text-Gen-Interface models on Baseten |
| OpenRouter | Supports all the text/chat/vision models from OpenRouter |
| Custom API Server (OpenAI Format) | Allows custom endpoint in OpenAI ChatCompletion format |
| Petals |
To get started with Open Interpreter, you first need to install it via
pippython --version
It is recommended to use a virtual environment to manage your dependencies. Once your environment is set up, you can install Open Interpreter with the following command:
pip install open-interpreter
Open Interpreter operates similarly to ChatGPT but runs locally on your machine. To start a new interactive chat, open the terminal and run:
interpreter
In Python, Open Interpreter retains the conversation history within a session. To start a fresh chat, you can reset the conversation history:
interpreter.messages = []
Open Interpreter allows you to save and restore conversations, ensuring you can pick up where you left off. In the terminal, conversations are saved in the
<your application directory>/Open Interpreter/conversations/interpreter --conversations
In Python, you can save the chat messages to a list and restore them later:
messages = interpreter.chat("Create a Python function…")
To reset the interpreter:
interpreter.messages = []
To resume the chat from the messages lists, use:
interpreter.messages = messages
You can customize the system message to modify permissions or provide additional context. In the terminal, this is done by editing the configuration file as described in the documentation. In Python, you can adjust the system message directly:
interpreter.system_message += """ Run shell commands with -y so the user doesn't have to confirm them. """ print(interpreter.system_message)
Open Interpreter uses LiteLLM to interface with different language models. You can switch models by setting the model parameter. In the terminal, use this syntax. For a full list of supported models, see Which LLMs Can You Use above.
interpreter --model gpt-4-turbo
In Python, you can set the model directly on the interpreter object:
interpreter.llm.model = "gpt-4-turbo"
You can run code directly through Open Interpreter. The computing environment is separate from the interpreter’s core, allowing independent execution. Here’s how to run a simple Python command:
from interpreter import interpreter interpreter.computer.run("python", "print('Welcome!')")
You can also prepare the environment by defining functions, setting variables, or logging into services before executing code:
interpreter.computer.run("python", "import replicate\nreplicate.api_key='...'") interpreter.custom_instructions = "Replicate is already imported." interpreter.chat("Please create a new image with Replicate...")
To get started building your LLM applications, check out GPTScript, Acorn’s framework that allows LLMs to operate and interact with various systems using natural language prompts.