aitelier 🎨🤖
An atelier for AI Agents running on your Mac thanks to mlx-llm.
How to install it
pip install aitelier
Why aitelier?
aitelier is not just another agent library — it’s my personal playground to learn how AI agents work
-
Learning by doing 🛠️: I want to understand every part of building AI agents. Instead of just using existing frameworks, I’m creating this library to figure out how things work from the ground up.
-
Exploration, not duplication 🧐:
aitelierisn’t meant to compete with other agent libraries. My goal is to learn, not to replace. Each update happens when I discover a new concept or method I want to explore. -
Deep understanding through practice 🧠: building from scratch helps me go beyond the surface. By recreating different paradigms, I can focus on truly understanding the details and improving my AI development skills.
aitelier will grow as I dive into new ideas and challenge myself to learn more. 🎯
I wrote an article about the motivation behind aitelier and how I implemented the first version of the library. You can read it here.
Agents as FSMs
In aitelier, agents are implemented as Finite State Machines (FSMs). Each agent has a set of states and transitions that define how the agent processes the input and produces the output.
Currently, aitelier supports two types of agents:
- Agent: A basic agent that takes a query and returns a response based on the available tools
- ReAct Agent: An agent the follows the Reasoning and Act paradigm paper with available tools
Agent Example
from aitelier.model import LLM
from aitelier.agents import Agent
from aitelier.tool import Tool
@Tool
def multiply(a: float, b: float) -> float:
return a * b
model = LLM("llama_3_2_3b_instruct")
agent = Agent(model=model, tools=[multiply])
agent("What is 3 multiplied by 4?")
ReAct Agent
The ReAct agent is a more advanced agent that follows the Reasoning and Act paradigm. The agent goes through the Think, Act, Observe states to produce the final answer based on a set of tools.
This is how you can use the ReAct Agent in aitelier:
from aitelier.model import LLM
from aitelier.agents import ReActAgent
from aitelier.tool import Tool
@Tool
def divide(a: float, b: float) -> float:
if b == 0:
return "Division by zero is not allowed"
return a / b
model = LLM("deepseek_r1_distill_llama_8b")
agent = ReActAgent(model=model, tools=[divide])
agent("What is 10 divided by 2?")
Supported Models
aitelier natively supports mlx-llm and Claude (with own API key)
Examples
Known Issues
aitelieris in active development, so expect bugs and breaking changes- The library is currently only tested on macOS
- All the agents depend on the LLM model - LLMs with fewer parameters are way more buggy than the larger ones since their behavior is less predictable - my advice is to use the largest models available
Metadata
Release files for aitelier 0.2.1
For a detailed explanation of source distributions (sdists) and built distributions (wheels), please see the package formats documentation.
Source distribution (sdist)
| File | Size | Uploaded | Provenance | Actions |
|---|---|---|---|---|
| aitelier-0.2.1.tar.gz | 17.3 kB | Without provenance | Details for aitelier-0.2.1.tar.gz |
Built distribution (wheel)
| File | Interpreter | ABI | Platform | Actions Reset |
|---|---|---|---|---|
| aitelier-0.2.1-py3-none-any.whl | Python 3 | none | any | Details for aitelier-0.2.1-py3-none-any.whl |
Total release size: 38.4 kB
Release files / aitelier-0.2.1.tar.gz
| Download URL | aitelier-0.2.1.tar.gz |
|---|---|
| Size | 17.3 kB |
| Tags | Source |
|
SHA-256 checksum How to use checksums |
a4fc634a960ab3a32f396b0c92048eb2ec8180703e054eb74523e0733ad5bc8c
|
|
BLAKE2b-256 checksum How to use checksums |
1b3c3c3ef97fdb0396c265ef2754b0a38e80092fdb326b93deabe54ca8278264
|
| Upload date | |
|
Uploaded using Trusted Publishing? What is trusted publishing? |
No |
| Uploaded via |
twine/6.1.0 CPython/3.10.6
|
Release files / aitelier-0.2.1-py3-none-any.whl
| Download URL | aitelier-0.2.1-py3-none-any.whl |
|---|---|
| Size | 21.1 kB |
| Tags | Python 3 |
|
SHA-256 checksum How to use checksums |
9ae2328bfed7442a05848cd8b416a841820c31bcfba45ea225af2b1bfdbd1519
|
|
BLAKE2b-256 checksum How to use checksums |
ee18587299ec92d2e391023d9a339ddecfc4e57e3e3a1ef89f1da9a854673d4c
|
| Upload date | |
|
Uploaded using Trusted Publishing? What is trusted publishing? |
No |
| Uploaded via |
twine/6.1.0 CPython/3.10.6
|
Security for aitelier 0.2.1
Provenance
Provenance describes where a file came from. On PyPI, provenance is shared via attestations, which provide a verifiable record of the build or publishing details. View details, limitations and caveats.
No provenance found
None of the files in this release were uploaded with provenance metadata.
A note on package security
PyPI is a repository for distributing packages, not a service for auditing them. While third-party security researchers and our staff do their best to detect and remove malicious packages, PyPI cannot guarantee the quality, security or suitability of any given package.
To learn more about protecting yourself from malicious actors, visit our package security documentation.