llm-registry 0.5.0

A centralized registry for discovering and managing LLM model capabilities. Track model features, costs, and limitations across providers like OpenAI and Anthropic. Supports both verified model definitions and user-managed entries with local storage.

LLM Registry

A Python package for managing LLM model capabilities across multiple providers. Track pricing, features, and limitations in one place.

Why LLM Registry?

Working with multiple LLM providers means juggling different pricing models, feature sets, and API parameters. LLM Registry solves this by providing a single source of truth for model metadata.

Check a model's capabilities before making API calls, compare costs across providers, and maintain your own custom model configurations alongside the built-in registry.

Features

• Unified Interface: Query model capabilities across OpenAI, Anthropic, Google, Cohere, Mistral, Meta, and more
• Cost Tracking: Input/output token costs, cached token pricing, and context window limits
• Feature Discovery: Check support for streaming, tools, vision, JSON mode, and system prompts
• Dual Storage: Built-in registry with 100+ models, plus local storage for custom configurations
• Multi-Provider Support: Single model can be associated with multiple providers
• CLI & Python API: Use from the command line or integrate into your code

Installation

pip install llm-registry

Quick Start

Python API

List all available models:

from llm_registry import CapabilityRegistry, Provider
registry = CapabilityRegistry()
models = registry.get_models()
for model in models:
    print(model)

Check a specific model's capabilities:

model = registry.get_model("gpt-5")
if model.api_params.stream:
    # Model supports streaming
    print(f"Cost: ${model.token_costs.input_cost}/M input tokens")

Add custom models:

from llm_registry import CapabilityRepository, Provider
from llm_registry.utils import create_model_capability

new_model = create_model_capability(
    model_id="gpt-5",
    provider=Provider.OPENAI,
    model_family="GPT-5",
    input_cost=0.01,
    output_cost=0.03,
    context_window=8192,
    supports_streaming=True,
    supports_tools=True
)

repo = CapabilityRepository()
repo.save_model_capabilities(new_model)

Command Line Interface

List models:

llmr list
llmr list --provider openai

Get detailed information:

llmr get gpt-5
llmr get gpt-5 --json

Example output:

{
  "model_id": "o1",
  "providers": ["openai"],
  "model_family": "o1",
  "api_params": {
    "max_tokens": true,
    "temperature": false,
    "top_p": false,
    "stream": true
  },
  "features": {
    "vision": true,
    "tools": true,
    "json_mode": true,
    "system_prompt": false
  },
  "token_costs": {
    "input_cost": 15.0,
    "output_cost": 60.0,
    "cache_input_cost": 7.5,
    "context_window": 200000,
    "training_cutoff": "2023-10"
  }
}

Manage custom models:

# Add a model
llmr add custom-model \
    --provider openai \
    --input-cost 0.01 \
    --output-cost 0.03 \
    --stream \
    --tools

# Update a model
llmr update custom-model --input-cost 0.005

# Delete a model
llmr delete custom-model --provider openai

Model Metadata

Each model tracks:

• Basic Information: Provider(s), model ID, model family
• Costs: Input/output token costs (per 1M tokens), cached token pricing
• API Parameters: max_tokens, temperature, top_p, streaming support
• Features: Vision, tools/function calling, JSON mode, system prompts
• Limits: Context window size, training data cutoff date

Configuration

Model data is stored in ~/.llm-registry by default. Override with:

repo = CapabilityRepository(data_dir=Path("/custom/path"))

Or via CLI:

llmr list --data-dir /custom/path

Contributing

Contributions welcome! Report bugs, suggest features, or submit pull requests.

License

MIT License. See LICENSE for details.