Tools
Standardcompute
By OpenClaw provider plugin for Standard Compute, an OpenAI-compatible LLM gateway with a fixed monthly budget
Install
openclaw plugins install clawhub:@henrikdukefoss/openclaw-standardcomputeREADME
# openclaw-standardcompute
[](https://github.com/henrikdukefoss/openclaw-standardcompute/actions/workflows/test.yml)
[OpenClaw](https://github.com/openclaw/openclaw) provider plugin for [Standard Compute](https://standardcompute.com), an OpenAI-compatible LLM gateway for AI agents.
Standard Compute plans are monthly with a fixed usage budget, starting at $19. Requests draw from that budget instead of being billed per token, and there are no automatic overage charges: when the budget runs out, requests return `402` until it renews or you add budget. That suits always-on OpenClaw agents whose token use is hard to predict. See [pricing](https://standardcompute.com/pricing) and [how budgets work](https://docs.standardcompute.com/budgets-and-limits).
This plugin is published by the Standard Compute team.
## Install
```bash
openclaw plugins install clawhub:@henrikdukefoss/openclaw-standardcompute
```
Requires OpenClaw 2026.10.3 or newer.
## Configure
Create an API key in the [dashboard](https://standardcompute.com/dashboard/api-keys), then either export it:
```bash
export STANDARDCOMPUTE_API_KEY=your-key
```
or save it through onboarding (**Standard Compute API key**):
```bash
openclaw onboard --auth-choice standardcompute-api-key
```
In scripts, pass the key with `--standardcompute-api-key <key>`.
`STANDARD_COMPUTE_API_KEY`, the name used by the `standardcompute` launcher, is also detected.
Make smart routing the default model:
```bash
openclaw models set standardcompute/standardcompute
openclaw config validate
```
- Base URL: `https://api.stdcmpt.com/v1`
- API: OpenAI Responses (`openai-responses`)
- Default model: `standardcompute/standardcompute`
## Models
| Model ref | What it does |
| --- | --- |
| `standardcompute/standardcompute` | Smart routing. The gateway picks a model per request from the models enabled on your [Models page](https://standardcompute.com/dashboard/models). Simple work goes to cheaper models, which stretches the budget. |
| `standardcompute/anthropic/claude-sonnet-5.5` | Pinned to Claude Sonnet 5.5 |
| `standardcompute/openai/gpt-5.6-sol` | Pinned to GPT-5.6 Sol |
| `standardcompute/deepseek/deepseek-v4-pro` | Pinned to DeepSeek V4 Pro |
| `standardcompute/z-ai/glm-5.3` | Pinned to GLM 5.3 |
| `standardcompute/moonshotai/kimi-k3` | Pinned to Kimi K3 |
| `standardcompute/qwen/qwen3.7-plus` | Pinned to Qwen 3.7 Plus |
Any other ID from the public model list also works, for example `standardcompute/anthropic/claude-opus-5.5` or a region-pinned `:eu` / `:us` variant:
```bash
curl -s https://api.stdcmpt.com/v1/models
```
A pinned model is served by that model only; the gateway never switches it for a different one. Pinned premium models use the budget faster than smart routing. Details: [models and routing](https://docs.standardcompute.com/models), [regions and privacy](https://docs.standardcompute.com/regions-and-privacy).
Model limits in the catalog are conservative (8,192 output tokens per request). To change them, add an entry under `models.providers.standardcompute.models` in your OpenClaw config.
## Other ways to connect
- The `standardcompute launch openclaw` launcher starts OpenClaw with a separate Standard Compute profile and leaves your usual setup alone. See the [OpenClaw guide](https://docs.standardcompute.com/agents/openclaw).
- A manual `models.providers` config block is in the same guide.
## Development
Requires Node 24.16+ (or 26.1+).
```bash
npm ci --ignore-scripts
npm test # builds dist/ and runs the tests; no network access
npm pack --dry-run
```
Built against OpenClaw `2026.10.3` (`openclaw.compat` and `openclaw.build` in `package.json`).
## Support
Open an issue in this repository, or see [docs.standardcompute.com](https://docs.standardcompute.com).
## License
MIT
tools
Comments
Sign in to leave a comment