AI Bricks

AI Bricks provide minimalistic toolbox for creating Generative AI based systems.

Features:

• unified interface to multiple Generative AI providers
• strong support for local model servers (KoboldCpp, LMStudio, Ollama, LlamaCpp, tabbyAPI, ...)
• ability to record requests and responses (in sqlite) and generate reports (ie usage)
• configuration driven approach to prompt templates (yaml + jinja2)
• minimal dependencies (requests, yamja)

AI Bricks focuses on providing basic building blocks rather than a complex framework. This aligns with research showing that the most successful LLM implementations use simple, composable patterns rather than complex frameworks. By keeping things minimal, it allows developers to build exactly what they need without unnecessary abstractions.

Examples

Example from the aisuite repo:

import aibricks
client = aibricks.client()

models = ["openai:gpt-4o", "xai:grok-beta"]

messages = [
    {"role": "system", "content": "Respond in Pirate English."},
    {"role": "user", "content": "Tell me a joke."},
]

for model in models:
    response = client.chat(
        model=model,
        messages=messages,
        temperature=0.75
    )
    print(response['choices'][0]['message']['content'])

Common Vision API

import aibricks
client = aibricks.client()

models = [
    "openai:gpt-4o",
    "xai:grok-vision-beta",
    "anthropic:claude-3-5-sonnet-latest",
    "lmstudio:moondream2",
]

messages = [
    {"role": "user", "content": [
        {"type": "text", "text": "what these two images have in common?"},
        {"type": "image_url", "image_url": {"url": "https://example.com/image1.jpg"}},
        {"type": "image_url", "image_url": {"url": "file://path/to/image2.jpg",
                                            "detail": "high"}},
    ]},
]

for model in models:
    response = client.chat(
        model=model,
        messages=messages,
    )
    print(response)

Minimalistic implementation of MemGPT-like agent

This example can be found in the examples/memgpt directory.

Supported providers

Provider	Example Connection String	Environmental Variables	Notes
OpenAI	openai:gpt-4o-mini	OPENAI_API_KEY
Anthropic	anthropic:claude-3-5-haiku-latest	ANTHROPIC_API_KEY
XAI	xai:grok-beta	XAI_API_KEY
Google	google:gemini-1.5-flash	GEMINI_API_KEY
DeepSeek	deepseek:deepseek-chat	DEEPSEEK_API_KEY
OpenRouter	openrouter:openai/gpt-4o	OPENROUTER_API_KEY
ArliAI	arliai:Llama-3.1-70B-Tulu-2	ARLIAI_API_KEY
Together	together:google/gemma-2b-it	TOGETHER_API_KEY	🚧🚧🚧
HuggingFace	huggingface:meta-llama/Meta-Llama-3-8B-Instruct-Turbo	HUGGINGFACE_API_KEY	🚧🚧🚧
Ollama	ollama:qwen2.5-coder:7b	-	GGUF
LMStudio	lmstudio:qwen2.5-14b-instruct	-	GGUF dynamic model loading
KoboldCpp	koboldcpp	-	GGUF
LlamaCpp	llamacpp	-	GGUF
tabbyAPI	tabbyapi	TABBYAPI_API_KEY	EXL2, GPTQ
GPT4All	gpt4all:Reasoner v1	-	GGUF buggy
vLLM	vllm:/opt/models/qwen2.5-coder-3b-instruct-q4_0.gguf	-	GGUF
dummy	dummy	-

License

MIT

Installation

Don't. The project is still in its infancy.

References

• Building Effective Agents
• Building an AI-Powered Game
• LLMs as Operating Systems: Agent Memory
• Letta (formerly MemGPT)
• MemGPT: Towards LLMs as Operating Systems
• Cognitive load is what matters
• aisuite repo
• ai-bricks-v3 repo

Chat API docs

• OpenAI Chat API
• Anthropic Messages API
• XAI Chat Completion
• Google Text Generation
• DeepSeek Docs
• OpenRouter Quick Start
• ArliAI Usage Docs
• Together Chat API
• HuggingFace API

Vision API docs

• OpenAI Vision API
• Anthropic Vision API

Local model servers

• Ollama
• LMStudio
• KoboldCpp
• LlamaCpp
• tabbyAPI
• GPT4All
• vLLM

Local model servers (TODO)

• LocalAI
• Fast-LLM
• TGI