Tools
ModelFetch
By An OpenClaw plugin to fetch, refresh, and synchronize available models from specified providers directly into the WebUI configuration.
Install
npm install
npmConfiguration Example
{
"plugins": {
"entries": {
"model-fetch": {
"config": {
"providers": {
"openrouter": {
"filter": {
"includeRegex": "^(openai|anthropic)/",
"excludeRegex": "(embed|rerank)",
"manualIds": ["openai/gpt-4o-mini"],
"replace": false
}
},
"ollama": {
"filter": {
"manualIds": ["qwen3:latest", "llama3.1:8b"]
}
}
}
}
}
}
}
}README
# OpenClaw Model Fetch
OpenClaw tool plugin for refreshing a configured model provider's model list and writing the selected models to `models.providers.<providerId>.models`, where the WebUI can select them.
Each provider has an independent persisted filter strategy. The same `model_fetch` tool can refresh OpenRouter with one regex and Ollama with another manual selection without the two providers sharing settings.
## Features
- Fetches model lists from:
- OpenAI-compatible providers (`openai-completions`, `openai-responses`, `openai-chatgpt-responses`)
- Anthropic-compatible providers (`anthropic-messages`)
- Ollama (`ollama`, using `/api/tags`)
- OpenRouter (`openrouter`, using `https://openrouter.ai/api/v1/models`)
- Persists selected models through OpenClaw's config mutation API with automatic post-write reload.
- Supports include regex, exclude regex, and exact manual model IDs per provider.
- Supports one-call refresh using only the provider's persisted strategy.
- Supports temporary filters, `replace`, and `dryRun` in tool calls.
- Preserves existing model entries when an ID already exists, including custom names, costs, modalities, and other model fields.
- Does not expose API keys in tool results.
- Uses OpenClaw's SSRF-guarded fetch and bounded response reader.
## Requirements
- Node.js 24.16 or newer
- OpenClaw with plugin API support (`2026.5.17` or newer; validated with OpenClaw `2026.10.5`)
- A provider already configured under `models.providers`
- Provider authentication configured through OpenClaw's normal model auth flow
## Installation
```bash
npm install
npm run plugin:build
npm run plugin:validate
```
Install from this checkout:
```bash
openclaw plugins install /absolute/path/to/OpenClaw-ModelFetch
```
For local development, link the checkout instead of copying it:
```bash
openclaw plugins install /absolute/path/to/OpenClaw-ModelFetch --link
```
Use `openclaw plugins list` and `openclaw plugins doctor` to confirm the plugin is discovered and healthy.
## Provider Filter Configuration
Filters live in the `model-fetch` plugin config and are keyed by provider ID:
```json
{
"plugins": {
"entries": {
"model-fetch": {
"config": {
"providers": {
"openrouter": {
"filter": {
"includeRegex": "^(openai|anthropic)/",
"excludeRegex": "(embed|rerank)",
"manualIds": ["openai/gpt-4o-mini"],
"replace": false
}
},
"ollama": {
"filter": {
"manualIds": ["qwen3:latest", "llama3.1:8b"]
}
}
}
}
}
}
}
}
```
Filter semantics:
- `includeRegex`: when set, only fetched IDs matching this JavaScript regex are selected.
- `excludeRegex`: matching IDs are removed from the regex-selected set.
- `manualIds`: exact IDs to select. When `includeRegex` is not set, only these IDs are selected. When `includeRegex` is set, these IDs bypass include/exclude matching. A manual ID is retained during `replace` if it already exists in config.
- `replace`: when `false` or omitted, existing models are preserved and new selected models are added. When `true`, models not in the selected fetched set and not manually selected are removed.
- Invalid regexes fail the tool call with a clear error.
Temporary tool arguments override the persisted values for that call only. Omitted arguments fall back to the provider's persisted strategy.
## Tool Usage
Refresh a provider with its already-persisted strategy:
```json
{
"provider": "openrouter"
}
```
Preview a temporary regex without changing config:
```json
{
"provider": "openrouter",
"includeRegex": "anthropic/",
"dryRun": true
}
```
Use a temporary manual selection:
```json
{
"provider": "ollama",
"manualIds": ["qwen3:latest", "llama3.1:8b"]
}
```
Synchronize the configured model list and remove stale entries:
```json
{
"provider": "ollama",
"replace": true
}
```
The result reports the provider, fetched count, selected model IDs, added/unchanged/removed counts, whether `replace` was active, whether it was a dry run, the config path, and whether the config was persisted.
## Provider Prerequisites
Example:
```json
{
"models": {
"providers": {
"openrouter": {
"baseUrl": "https://openrouter.ai/api/v1",
"api": "openai-responses",
"models": []
}
}
}
}
```
For Ollama or another private/local endpoint, explicitly opt in to private-network requests:
```json
{
"models": {
"providers": {
"ollama": {
"baseUrl": "http://127.0.0.1:11434",
"api": "ollama",
"request": {
"allowPrivateNetwork": true
},
"models": []
}
}
}
}
```
## Development
```bash
npm test
npm run plugin:build
npm run plugin:validate
```
The generated `openclaw.plugin.json` should list `model_fetch` under `contracts.tools`.
tools
Comments
Sign in to leave a comment