llms-dev 0.1.0

Fuzzy search LLM models from models.dev

llms

Fuzzy search LLM models from models.dev in your terminal.

日本語版 README はこちら

Demo

$ llms
Model> claude-sonnet-4.6

  anthropic/claude-sonnet-4-6          | Claude Sonnet 4.6
  opencode/claude-sonnet-4-6           |
  venice/claude-sonnet-4-6             | Provider:    Anthropic
  ...                                  | Model ID:    anthropic/claude-sonnet-4-6
                                       | Cost:        $3.00 / $15.00
                                       | Context:     200K

Copied: anthropic/claude-sonnet-4-6

Installation

Prerequisites

• Python 3.13+
• fzf (for interactive mode)

Install

# uv (recommended)
uv tool install -e .

# pipx
pipx install -e .

# pip
pip install -e .

Or run ./install.sh to auto-detect the best available installer.

Install from GitHub (without cloning)

uv tool install git+https://github.com/atsuya.sakata/llm-models.git
pipx install git+https://github.com/atsuya.sakata/llm-models.git

This makes the llms command available globally.

Usage

Interactive Mode (fzf)

llms                  # Fuzzy search → Enter to copy model ID to clipboard
llms --refresh        # Force refresh cache
llms --no-copy        # Print model ID to stdout (for piping)
llms --json           # Output selected model details as JSON
llms -p anthropic     # Filter by provider

Key Bindings

Key	Action
Type	Fuzzy search
Up/Down	Navigate models
Enter	Copy model ID to clipboard
Ctrl-C	Exit

Programmatic Mode (for AI agents & scripts)

Non-interactive subcommands that output structured data — no fzf required.

# Get a model by ID
llms get anthropic/claude-sonnet-4-6 --json

# List models with filters
llms list --json
llms list --cap tool_call --min-context 128k --sort cost.input --limit 5 --json
llms list --cap reasoning --max-input-cost 5.0 --json

# Text search
llms search claude --json --limit 5
llms search llama -p meta --json

# List providers
llms providers --json

Filter Flags (for list and search)

Flag	Description	Example
-p, --provider	Filter by provider	-p anthropic
--cap	Capability filter (repeatable)	--cap tool_call --cap reasoning
--min-context	Minimum context window	--min-context 128k
--max-input-cost	Max input cost ($/1M tokens)	--max-input-cost 3.0
--sort	Sort by field	--sort cost.input
--limit	Max number of results	--limit 10

Output Formats

Flag	Format
--json	Pretty-printed JSON
--jsonl	One JSON object per line
(none)	One model ID per line

Python Library

from llms.client import Client
from llms.query import Query

client = Client()

# Get by ID
model = client.get("anthropic/claude-sonnet-4-6")

# List with filters
models = client.list(Query(
    caps=["tool_call"],
    min_context=128_000,
    sort="cost.input",
    limit=5,
))

# Text search
results = client.search("claude")

# Providers
providers = client.providers()

Data Source

Model data is fetched from the models.dev /api.json endpoint. Responses are cached locally for 1 hour at ~/Library/Caches/llms/.

Development

uv sync
uv run pytest -v
uv run ruff check . && uv run ruff format --check .