Ollama · LM Studio · llama.cpp

One interface for every local LLM runtime

Ollama, LM Studio, and llama.cpp behind a single SDK and CLI. Write your call once; Loomcast routes it to the right runtime.

$pip install loomcast
Get started →
loomcast
$ loomcast compare "Explain recursion in one sentence."

┌── OLLAMA (llama3) ──┐ ┌─ LMSTUDIO (qwen) ─┐ ┌─ LLAMACPP (gemma) ─┐
│ A function that     │ │ When a function   │ │ Calling a func-    │
│ calls itself...     │ │ calls itself...   │ │ tion from itself.  │
└─────────────────────┘ └───────────────────┘ └────────────────────┘
Key Features

Everything you need to run local models

A thin, dependable layer over the runtimes you already run — not a server, not an inference engine.

Explore all actions →

Unified Interface

Write manager.chat(model, messages) once. Loomcast routes it to the right runtime — no per-runtime ports, routes, or payload formats to learn.

Three Local Runtimes

Ollama, LM Studio, and llama.cpp supported from day one, each behind the same RuntimePort contract.

Catalog & Recommendations

Search and filter models by task (code, math, vision) with a dynamic catalog and offline static fallback.

SHA-256 Verification

Verify installed model integrity against SHA-256 digests and update models seamlessly.

Resilient by Design

The dynamic catalog automatically falls back to a checksum-verified static catalog when the network misbehaves.

Side-by-Side Comparison

Send one prompt to every runtime at once and compare responses in parallel panels.

CLI and SDK

Drive it from the terminal with loomcast ..., or import loomcast and call the same operations from Python.

Clean Architecture

Ports and adapters throughout: the domain never imports the network. Add a new runtime in a single file.

Interactive Explorer

LoomCast Features & Actions Reference

Explore all 13 operations available across the CLI terminal interface and the LoomCast Python SDK.

Inference & Comparison

Side-by-Side Comparison

Sends the same prompt concurrently to Ollama, LM Studio, and llama.cpp, displaying outputs side-by-side in parallel panels to evaluate speed and output quality.

CLI Command Syntax
loomcast compare "Explain recursion in one sentence."

Options & Arguments

Option / FlagTypeDescription
promptargumentPrompt sent to all selected runtimes
--model, -moptionForce specific model across all runtimes
--backend, -brepeatableRestrict comparison to specific backends

Simulated Terminal Output

loomcast execution output
Querying runtimes...

┌── OLLAMA (llama3) ────────┐ ┌── LMSTUDIO (qwen) ────────┐ ┌── LLAMACPP (gemma) ───────┐
│ A function calling itself │ │ When a program function   │ │ Self-referencing function  │
│ until reaching a base     │ │ repeatedly invokes itself │ │ execution until base step  │
│ condition.                │ │ to solve sub-problems.    │ │ is reached.                │
└───────────────────────────┘ └───────────────────────────┘ └────────────────────────────┘
Benchmarking Playground

Multi-Runtime Comparison Simulator

The loomcast compare command queries all your active local runtimes in parallel. Test the interactive preview below:

Preset Prompts:
OLLAMA
llama3:8b

Prompt sent

"Explain recursion in one sentence."

Generated response

Recursion is a programming concept where a function calls itself until reaching a base condition that stops execution.
Status: 200 OKTime: 1.2s
LM STUDIO
qwen2.5-7b

Prompt sent

"Explain recursion in one sentence."

Generated response

It occurs when a subprogram repeatedly invokes itself to break a complex problem into identical smaller subproblems.
Status: 200 OKTime: 0.9s
LLAMA.CPP
gemma-2b

Prompt sent

"Explain recursion in one sentence."

Generated response

A function is recursive when its body directly or indirectly calls itself until a termination base case.
Status: 200 OKTime: 0.6s
System Integrity & Resiliency

Automated System Diagnostics

LoomCast includes an automated self-diagnostic suite. Validate your configuration, runtime HTTP connectivity, and catalog integrity in a single command.

✓Configuration file audit (~/.config/loomcast/config.json)
✓Environment variable inspection (LOOMCAST_*)
✓HTTP connectivity test on every runtime port
✓Dynamic catalog access check with automatic static fallback
✓SHA-256 cryptographic verification of embedded static catalog
loomcast diagnostics
$ loomcast doctor

Loomcast System Diagnostics

✓ Configuration: loaded successfully from C:\Users\User\.config\loomcast\config.json
✓ Environment variables: none set, using defaults
✓ Runtime OLLAMA: available (http://127.0.0.1:11434)
✓ Runtime LMSTUDIO: available (http://127.0.0.1:1234)
⚠ Runtime LLAMACPP: unavailable (not default backend)
✓ Catalog: dynamic catalog reachable
✓ Static catalog integrity: SHA-256 check passed

0 problem(s) found, 1 warning(s)

The same operations, two ways

Everything the CLI does is available from the Python SDK, backed by the same manager.

Command line
# Check which runtimes are up
loomcast status

# Send one prompt to every runtime, side by side
loomcast compare "Explain recursion in one sentence."

# Chat with a specific model on a specific backend
loomcast chat llama3 "Hello world!" --backend ollama
Python SDK
import loomcast

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

# Send a chat prompt
reply = loomcast.chat(
    "llama3",
    [{"role": "user", "content": "Hello!"}],
    runtime_type="ollama",
)

# Compare the same prompt across runtimes
for r in loomcast.compare("Explain recursion in one sentence."):
    print(r.runtime_type.value, "->", r.response or r.error)