Loomcast

Python SDK and CLI for seamlessly managing local LLM runtimes (Ollama, LM Studio, and llama.cpp).

Installation

pip install loomcast

To work on Loomcast itself, install it from a clone in editable mode with the development dependencies:

pip install -e ".[dev]"

Quick Start

CLI Usage

# Check status of local runtimes
loomcast status

# Diagnose configuration, runtimes, catalog and integrity
loomcast doctor

# List installed models
loomcast list

# Search catalog for models
loomcast search code

# Recommend models for a specific task or tag
loomcast recommend coding

# Install a model
loomcast install llama3 --backend ollama

# Batch install all models in a category
loomcast install-category code --backend ollama

# Send a prompt to a model
loomcast chat llama3 "Hello world!" --backend ollama

# Verify installed model integrity
loomcast verify llama3 --backend ollama

# Update an installed model
loomcast update llama3 --backend ollama

# Remove an installed model
loomcast remove llama3 --backend ollama

# View or change configuration
loomcast config
loomcast config --set-default lmstudio

Model installation, removal and updating are driven through the runtime's API, which only Ollama exposes. LM Studio and llama.cpp report an explanatory error instead: load models through the LM Studio application, or start llama-server with the model you want.

Python SDK Usage

import loomcast

print(loomcast.__version__)

# Search catalog
results = loomcast.search("code")
for model in results:
    print(model.name, model.description)

# Recommend models for a task
recs = loomcast.recommend("coding")

# List what is installed locally
print(loomcast.list_installed())

# Install a model, and batch install a whole category
loomcast.install("llama3", runtime_type="ollama")
installed, failed = loomcast.install_category("code", runtime_type="ollama")

# Send chat messages
response = loomcast.chat(
    "llama3", [{"role": "user", "content": "Hello!"}], runtime_type="ollama"
)
print(response)

# Verify model integrity
verification = loomcast.verify("llama3", runtime_type="ollama")
print(f"Verified: {verification.verified}, Digest: {verification.digest}")

# Update returns True when the digest changed, False when already current
changed = loomcast.update("llama3", runtime_type="ollama")

# Remove an installed model
loomcast.remove("llama3", runtime_type="ollama")

For finer control over configuration or dependency injection, drive LoomcastManager directly:

from loomcast import LoomcastConfig, LoomcastManager

manager = LoomcastManager(config=LoomcastConfig(default_backend="ollama"))
print(manager.list_available_runtimes())

Documentation

Detailed documentation guides are available in the docs/ directory:

• Getting started
• Configuration
• Supported runtimes
• Extending & architecture
• CLI reference