local-model

Portable local-agent helpers for using a LAN Ollama server with Codex, Claude Code, and OpenCode.

What This Installs

• local-model: switch and inspect models exposed by an Ollama server.
• codex-local: run Codex against the selected local model.
• claude-local: run Claude Code against the selected local model.
• opencode-local: run OpenCode against the selected local model.

Plain codex, claude, and opencode keep their normal cloud behavior.

Client Install

git clone git@github.com:YOUR_USER/local-model.git
cd local-model
bin/install-client.sh \
  --base-url http://YOUR_OLLAMA_HOST:11434 \
  --default-model gemma4:26b \
  --small-fast-model qwen3.5:4b
source ~/.bashrc

Usage

local-model current
local-model list
local-model caps qwen3.6-agent:35b
local-model use qwen3.6-agent:35b
local-model use-small qwen3.5:4b

codex-local
claude-local
opencode-local

Ollama Host Setup

Run this on the Ollama host to create tuned -agent aliases without changing base model tags:

ollama-host/create-agent-models.sh

Then switch from any installed client:

local-model use qwen3.6-agent:35b
local-model use-small qwen3.5:4b

Why Agent Aliases

Ollama tags can carry runtime defaults. Agent workflows need larger context, larger output budgets, and model-specific sampling. Separate -agent aliases let you keep base models unchanged while giving coding agents stable defaults.