llama-swap-sync

Automatically keep your opencode model list in sync with your llama-swap configuration — no manual editing required.

---

How it works

Whenever you save your config.yaml, a systemd path unit detects the change and runs a sync script that:

1. Parses all models from the models: block in your llama-swap config.yaml
2. Extracts each model's context size from -c / --ctx-size arguments
3. Derives the model display name from the .gguf filename
4. Adds new models, removes deleted ones, and updates context sizes in opencode.json
5. Backs up opencode.json before every change

config.yaml saved → systemd detects change → llama-swap-sync.sh runs → opencode.json updated

---

Requirements

• Linux with systemd (user units)
• python3 — stdlib only, no extra packages
• bash

---

Install

git clone https://github.com/janvitos/llama-swap-sync
cd llama-swap-sync
chmod +x install.sh
./install.sh

The installer will:

• Copy llama-swap-sync.sh to ~/.local/bin/
• Install the systemd units to ~/.config/systemd/user/
• Enable and start the file watcher
• Run first-time interactive setup, then do the initial sync

First-time setup

On first run you'll be prompted for:

Prompt	Default
Path to opencode.json	~/.config/opencode/opencode.json
Path to config.yaml	~/llama-swap/config.yaml
llama-swap baseURL	http://127.0.0.1:41234/v1
Provider ID	llama-swap
Provider display name	llama-swap

Settings are saved to ~/.config/llama-swap-sync/settings.conf.

---

Usage

Command	Description
(edit and save config.yaml)	Sync runs automatically
llama-swap-sync.sh	Run sync manually
llama-swap-sync.sh --reconfigure	Reset saved settings
llama-swap-sync.sh --help	Show usage

Useful systemd commands

# Check watcher status
systemctl --user status llama-swap-sync.path

# Follow sync logs live
journalctl --user -u llama-swap-sync.service -f

# View last sync output
journalctl --user -u llama-swap-sync.service -n 30

---

What gets written to opencode.json

Given this config.yaml:

models:
  "qwen3-coder":
    cmd: |
      llama-server --port ${PORT} -c 32768 \
        -m /models/Qwen3-Coder-30B-Q4_K_M.gguf
    proxy: http://127.0.0.1:${PORT}

  "gemma3-4b":
    cmd: llama-server --port ${PORT} -c 8192 -m /models/gemma-3-4b-it-Q8_0.gguf
    proxy: http://127.0.0.1:${PORT}

The script produces this provider block in opencode.json:

{
  "provider": {
    "llama-swap": {
      "npm": "@ai-sdk/openai-compatible",
      "name": "llama-swap",
      "options": {
        "baseURL": "http://127.0.0.1:41234/v1"
      },
      "models": {
        "qwen3-coder": {
          "name": "Qwen3-Coder-30B-Q4_K_M",
          "limit": { "context": 32768 }
        },
        "gemma3-4b": {
          "name": "gemma-3-4b-it-Q8_0",
          "limit": { "context": 8192 }
        }
      }
    }
  }
}

---

Notes

• JSONC supported — opencode.json files with // comments and trailing commas are handled correctly
• Backups — a timestamped .bak file is created next to opencode.json before every write
• Both write styles detected — the systemd path unit watches for both in-place writes and temp-file-rename writes (vim, nano, VS Code, etc.)
• No network calls — the script reads config.yaml directly; it does not query the llama-swap API

---

Files

File	Description
llama-swap-sync.sh	Main sync script
llama-swap-sync.path	systemd unit — watches config.yaml for changes
llama-swap-sync.service	systemd unit — runs the sync script
install.sh	Installs everything and runs first-time setup

---

License

MIT