Local-first AI coding CLI

Alpha Software — Use at Your Own Risk

Locode is under active development and has not been validated for production use. Interfaces, configuration formats, and behaviours may change without notice between releases. It is provided as-is, without warranty of any kind. Use in critical or production environments is not recommended at this stage.

What's New

Apr 3, 2026

v0.1.17

Recommend local model from eval results

Apr 3, 2026

v0.1.16

Add local model evaluation harness

Apr 3, 2026

v0.1.15

Finish v0.3.5 hardening

How It Works

1

You type

Write your prompt — a question, a task, a code request.

2

Locode routes

Smart classification decides which model handles your task.

3

Right model responds

Ollama for simple tasks (free). Claude for complex ones (powerful).

Features

Smart Routing

Regex rules + LLM-assisted classification. Configurable thresholds and patterns in locode.yaml.

Token Savings

Track per-agent and total token usage with USD cost estimates. Benchmark hybrid vs claude-only vs local-only.

Offline Ready

Works entirely offline without an API key. Falls back to Ollama automatically when Claude is unavailable.

MCP Support

Extend with Model Context Protocol servers — including remote ones like Linear.

Save on Every Session

Mode	Claude Tokens Used	Estimated Cost
Claude-only	100%	$$$
Hybrid (locode)	~40–60%	$
Local-only	0%	Free

Run locode benchmark to see your own savings.

Quick Start

# Install
npm install -g @chocks-dev/locode

# First-time setup
locode setup

# Start coding
locode