aimo-registry 0.1.3

Smart AI model registry with hierarchical access and rich metadata

aimo

Proof of concept: aimo is an experimental model-registry package. It is useful enough to try, but there is no clear long-term maintenance plan yet. Expect API changes, stale upstream metadata, and rough edges until the project proves its shape.

Smart AI Model Registry — Hierarchical access to AI model identifiers with rich metadata.

Combines metadata from LiteLLM and models.dev.

Features

• ✨ Hierarchical access: aimo.openai.gpt_4o, aimo.anthropic.claude_3_5_sonnet
• 🧠 Rich objects that act like strings: pass them to DSPy, LiteLLM, LangChain, SDKs, etc.
• 📊 Metadata: context window, pricing, capabilities, vision support, and upstream raw fields
• 🔄 Explicit cache refresh from LiteLLM + models.dev via aimo.update()
• 🚀 Zero runtime dependencies; lazy startup with local caching
• 🧩 Static .pyi stubs for LSP autocomplete of common aimo.openai.gpt_4o paths

Installation

pip install aimo-registry

Usage

import aimo

# Hierarchical access. The first registry access loads cached data, or downloads it
# if the cache has not been populated yet.
model = aimo.openai.gpt_4o
model = aimo.anthropic.claude_3_5_sonnet_20240620
model = aimo.groq.llama3_70b_8192

# Works as a string.
print(model)                    # "gpt-4o"
assert isinstance(model, str)

# Human-readable summary when you want one.
print(model.pretty())

# Common attributes from LiteLLM + models.dev are available when present.
print(model.context)            # e.g. 128000
print(model.input_price)        # e.g. 2.5e-06
print(model.output_price)       # e.g. 1e-05
print(model.supports_vision)    # e.g. True
print(model.capabilities)       # e.g. ['vision', 'function_calling']

# Full raw data.
print(list(model.raw.keys()))

# Use anywhere a string is expected.
# lm = dspy.LM(model)
# response = litellm.completion(model=model, messages=[...])

Search & Aliases

import aimo

# Search
models = aimo.search("claude")
models = aimo.search("gpt-4", provider="openai")
models = aimo.search(provider="groq")

# Popular aliases, when present in the upstream data
model = aimo.gpt4o
model = aimo.claude35
model = aimo.sonnet
model = aimo.haiku
model = aimo.o1
model = aimo.o3

Updating the Model List

import aimo

aimo.update()   # Force refresh from LiteLLM + models.dev and update the registry in place.

By default, aimo stores upstream JSON files in ~/.cache/aimo. Set AIMO_CACHE_DIR to use a different cache directory.

License

MIT