interfaces-to 0.2.30

The quickest way to make Large Language Models do things.

Add a little action to your LLM adventure with 🐙 Interfaces

🐙 Interfaces (aka into) is the quickest way to make Large Language Models do things. It's a simple, powerful and flexible way to build more useful, more engaging and more valuable agent-driven applications with LLMs.

✨ Key Features

⭐️ Built-in tools for common tasks and platforms (see all)
⭐️ Dynamic message sources for real-time interactions (see all)
⭐️ Start building with just 4 only 3(!) lines of code
⭐️ Create agents with system messages to control behaviour (more info)
⭐️ Beginner-friendly Python library, learn and teach coding with 🐙 Interfaces!
⭐️ Simple and secure configuration
⭐️ Fully compatible with the OpenAI API SDK
⭐️ Works with gpt-4o, gpt-4o-mini and other OpenAI models
⭐️ Works with llama3.1, mistral-large and more via Ollama
⭐️ Works with Azure OpenAI (more info)
⭐️ Supports (thrives) on parallel_tool_calls (more info)
⭐️ Works with any other LLM applications and services that support the OpenAI API
⭐️ Runs on your local machine, in the cloud, or on the edge
⭐️ Run tools from the command line with the into CLI (see all)
⭐️ Extensible design for building custom tools (example) and message sources (example)
⭐️ Open-source, MIT licensed, and community-driven

🚀 Quick Start

Installation

Install with pip:

pip install interfaces-to

or

Install with poetry:

poetry add interfaces-to

Usage

Turn your OpenAI completion into a fully capable agent with 3 lines of code:

# 1️⃣ import `into`
import interfaces_to as into

# 2️⃣ import the OpenAI client as normal
from openai import OpenAI
client = OpenAI()

# 3️⃣ add your favourite tools and set a message
agent = into.Agent().add_tools(['Slack','OpenAI']).add_messages("What was the last thing said in each slack channel? Write a 5 line poem to summarise and share it in an appropriate channel")

# 4️⃣ start the agent loop, with an OpenAI completion
while agent:
  agent.completion = client.chat.completions.create(
    model="gpt-4o",
    messages=agent.messages,
    tools=agent.tools,
    tool_choice="auto"
  )

# 5️⃣ watch the magic happen! 🎩✨

This prints the following output:

[user]		  What was the last thing said in each slack channel? Write a 5 line poem to summ
		    arise and share it in an appropriate channel

[assistant]     Calling 1 tool:
		    list_channels({})

[tool]	      Output of tool call list_channels({})
		      Channels: [{'id': 'C07EEUES770', 'name': 'general', 'is_channel':...

[assistant]	  Calling 7 tools:
                read_messages({"channel": "general"})
                read_messages({"channel": "hello"})
                read_messages({"channel": "fun-times"})
                read_messages({"channel": "poetry"})
                read_messages({"channel": "jokes"})
                read_messages({"channel": "welcome"})
                read_messages({"channel": "random"})

[tool]		  Output of tool call read_messages({"channel": "random"})
		    Messages: [{'subtype': 'channel_join', 'user': 'U07ET3LMDB7', ...

[tool]		  Output of tool call read_messages({"channel": "welcome"})
		    Error reading messages: not_in_channel

[tool]		  Output of tool call read_messages({"channel": "jokes"})
		    Messages: [{'user': 'U07EP9STUTU', 'type': 'message', 'ts': '1722477596.990159', ...

[tool]		  Output of tool call read_messages({"channel": "poetry"})
		    Messages: [{'user': 'U07EP9STUTU', 'type': 'message', 'ts':...

[tool]		  Output of tool call read_messages({"channel": "fun-times"})
		    Messages: [{'user': 'U07EP9STUTU', 'type': 'message', 'ts': ...

[tool]		  Output of tool call read_messages({"channel": "hello"})
		    Messages: [{'subtype': 'channel_join', 'user': 'U07EP9STUTU', 'text': '<@U07EP9STUTU> has joined the channel', 'type': 'message', 'ts': '1722421262.722359'}]

[tool]		  Output of tool call read_messages({"channel": "general"})
		    Messages: [{'user': 'U07EP9STUTU', 'type': 'message', 'ts': ...

[assistant]	  Here is a summary poem of the last messages shared in the Slack channels:
		
                In **#general**, they said with glee,
                "Atoms make up all that we see!"
                **#hello** greeted someone new,
                Joining the channel, the greetings flew.
                
                **#fun-times** loved a joke,
                On atoms that can't be broke.
                In the **#poetry** nook, atoms we praised,
                With playful rhymes, our spirits raised.
                
                **#jokes** was all about the spin,
                Electrons in a debate always win.
                And though **#welcome** remained hushed,
                **#random** laughed at atoms crushed.
                
                I'll share this poem in the **#poetry** channel.

[tool]		  Output of tool call send_slack_message({"channel":"poetry","message":"Here's a "...
                Message sent to poetry with timestamp 1722493789.651039: 
                Here's a summary poem of our last messages:
                
                In **#general**, they said with glee,
                "Atoms make up all that we see!"
                **#hello** greeted someone new,
                Joining the channel, the greetings flew.
                
                **#fun-times** loved a joke,
                On atoms that can't be broke.
                In the **#poetry** nook, atoms we praised,
                With playful rhymes, our spirits raised.
                
                **#jokes** was all about the spin,
                Electrons in a debate always win.
                And though **#welcome** remained hushed,
                **#random** laughed at atoms crushed.

[assistant]	  I have shared the poem summarizing the last messages in each channel
                to the **#poetry** channel.

agent.messages is also updated with the latest messages and retains the format needed by the OpenAI SDK, so you can continue the adventure and build more complex applications.

You can run this example in this Jupyter notebook.

Setting the system message

You can modify the behaviour of your agent by setting the system message.

agent = into.Agent("Always talk like a pirate")

Configuring tools

Using environment variables (Recommended for production)

Tools usually require a token. Tokens can always be configured by setting the relevant environment variables. e.g. for Slack you can set the SLACK_BOT_TOKEN environment variable.

If you are using environment variables, you can take advantage of agent.add_tools or the into.import_tools function to automatically configure your tools. This function will look for the relevant environment variables and configure the tools with default settings.

agent.add_tools(['Slack'])

or

tools = into.import_tools(['Slack'])

Using a .env file (Recommended for local development)

You can also configure your tools using a .env file. This is useful if you want to keep your tokens and other settings in a single file and helpful for local development.

Simply add a .env file in the root of your project with the following format:

SLACK_BOT_TOKEN=xoxb-12345678-xxxxxxxxxx

Setting tokens directly in code

If you prefer to set the token directly in your code or have more control over tool settings, you can do so by passing arguments to each tool. Tokens provided in code will override any environment variables.

You can optionally restrict functions to only those which you need.

Here's an example of configuring the Slack tool:

tools = [*into.Slack(
    token="xoxb-12345678-xxxxxxxxxx",
    functions=["send_slack_message"]
)]

Note that each tool is preceded by an asterisk * to unpack the tool's functions into a list, which the OpenAI API SDK expects.

📦 Available tools

into comes with loads of pre-built tools to help you get started quickly. These tools are designed to be simple, powerful and flexible, and can be used in any combination to create a wide range of applications.

Tool	Description	Functions	Configuration
Self	Encourage self awareness, time awareness and logical evaluation.	wait, plan, get_time, do_math	None required.
System	Control the system behaviour.	get_system_message, set_system_message, clear_system_message	None required.
OpenAI	Create completions and embeddings with the OpenAI API (Yes, that means self-prompting 🔥)	create_chat_completion, create_embedding	Uses OPENAI_API_KEY environment variable
Slack	Send messages to Slack channels, create channels, list channels, and read messages	send_slack_message, create_channel, list_channels, read_messages	Uses SLACK_BOT_TOKEN environment variable
Notion	Find, read and create pages in Notion	search_notion, query_notion_database, read_notion_page, create_notion_page	Uses NOTION_TOKEN environment variable. Databases must be explicitly shared with the integration.
Airtable	Find, read and create records in Airtable	list_all_bases, get_base, list_base_records, create_base_records	Uses AIRTABLE_TOKEN environment variable
People Data Labs	Find information about people and companies	find_person, find_company	Uses PDL_API_KEY environment variable

More tools are coming soon:

• Twilio
• GitHub
• Jira
• Discord
• and more!

See the 🛠️ Tools Project plan for more information on upcoming tools.

🌐 Experimental: Dynamic messages

into supports reading messages dynamically. The messages variable required by OpenAI SDK is a list of dictionaries, where each dictionary represents a message. Each message must have a role key with a value of either user or assistant, and a content key with the message content.

You can use agent.add_messages or into.read_messages to configure dynamic messages.

agent.add_messages(["What was the last thing said in each slack channel? Write a 5 line poem to summarise and share it in an appropriate channel"])

or

messages = into.read_messages(["Slack"])

The following sources are currently supported:

Source	Description	Configuration
Slack	Read messages from a Slack channel where your app is mentioned or in direct messages	Requires SLACK_APP_TOKEN and SLACK_BOT_TOKEN environment variable. Socket Mode must be enabled with the appropriate events.
Ngrok	Receive POST /message body using Ngrok. Useful for testing webhooks locally.	Requires NGROK_AUTHTOKEN environment variable.
FastAPI	Receive POST /message body on Port 8080 with FastAPI.	None required.
Gradio	Receive messages from Gradio's ChatInterface.	None required.
CLI	Read messages from the command line. For use in scripts executed on the command line or with running into itself (see below).	None required.

See the 💬 Messages Project plan for more information on upcoming tools.

Limitations

• Currently only one source can be configured at a time.
• History is not retained between the resolution of messages, however into is able to simulate message history by calling the Slack read_messages tool if equipped with into.import_tools(['Slack']).

📟 Experimental: CLI Support

into supports running tools from the command line. This is useful when CLI is the message source, allowing you to run as a standalone application.

By default this uses OpenAI and requires the OPENAI_API_KEY environment variable.

You can install into with the CLI support by running:

pipx ensurepath
pipx install interfaces_to

Since pipx installs packages in an isolated environment, you may need to add the dependencies for the tools you want to use. To do this for the Slack tool, you can do this at install time by running:

pipx install interfaces_to[slack]

If you want to use additional tools after install, you can install the dependencies with:

pipx inject interfaces_to ngrok

Then you can run into with the following command:

into --tools=Slack "What was the last thing said in each slack channel? Write a 5 line poem to summarise and share it in an appropriate channel"

or run it with poetry by cloning this repository:

poetry install

Then you can run into with the following command:

poetry run into --tools=Slack "What was the last thing said in each slack channel? Write a 5 line poem to summarise and share it in an appropriate channel"

You can also pipe messages to into, which will output JSON for manipulation in other tools like jq:

echo "What was the last thing said in each slack channel? Write a 5 line poem to summarise and share it in an appropriate channel" | into --tools=Slack

Usage

• --help - Show help message.
• --tools - A comma-separated list of tools to import. e.g. --tools=Slack,OpenAI
• --messages - A comma-separated list of message sources to import. e.g. --messages=Slack,CLI
• --model - The model to use for completions. e.g. --model=gpt-4o
• --api-key - The OpenAI API key to use for completions. e.g. --api-key=sk-12345678
• --endpoint - The endpoint to use for completions. e.g. --endpoint=https://myendpoint
• --azure - Use Azure functions for completions. e.g. --azure
• [message] - The message to send to the tools when --messages=CLI is set. This can passed in via stdin or as the last argument. When provided, into will run the tools and output the result as JSON to stdout.

Use with Azure OpenAI

You can use into with Azure OpenAI by setting the flags below.

into --tools=Slack --azure --endpoint=https://azure-endpoint --model=gpt-4o --api-key=sk-12345678 "summarise the last messages in each slack channel"

Use with open source models via Ollama

You can use into with open source models via Ollama by setting the flags below. Important: The model MUST support function calling. Full list of compatible models can be found here.

into --tools=Slack --endpoint=http://localhost:11434/v1 --model=llama3.1:8b "what is the time?"

📚 Documentation (coming soon!)

For more information, check out the detailed documentation.

💬 Community

Join the 🐙 Interfaces Slack to chat with other LLM adventurers, ask questions, and share your projects.

🤝 Contributors (coming soon!)

We welcome contributions from the community! Please see our contributing guide for more information.

Notable contributors and acknowledgements:

@blairhudson • 🐙

🫶 License

This project is licensed under the MIT License - see the LICENSE file for details.