How do you use Plugsky with the OpenAI Playground?

Key facts

Plugsky endpoint	https://api.plugsky.com/v1 (OpenAI-compatible)
Playground config	Set a custom base URL and an sk-live-… key; the docs list this path as tested
Models	Use ids from the catalogue of 30+ models, such as plugsky-micro, plugsky-lite and plugsky-pro
Chat parameters	temperature, top_p, max_tokens, stop and stream are supported
Structured options	JSON mode and tools work through the same endpoint
Free to try	2 free models on the free plan, no card required
Fallback tools	Plugsky dashboard playground and the OpenAI-compatible API tester
Product status	Live; image, audio and batch endpoints are coming soon

TL;DR

• A custom base URL is the only setup a compatible playground needs.
• Use a scoped development key, never a production admin key.
• Iterate prompts visually, then copy the same parameters into code.
• If the custom URL field is missing, use the dashboard or API tester.
• Playground calls come from the browser — treat the key as exposed to that machine.

How it works, step by step

1. Create a development API key in the Plugsky dashboard with narrow scopes.
2. Open the playground and find the provider or custom endpoint settings.
3. Set the base URL to https://api.plugsky.com/v1.
4. Paste the key and load the model list, then select plugsky-lite or plugsky-pro.
5. Tune temperature, max_tokens and stop while comparing outputs.
6. Reproduce the winning settings in your application code.
7. Rotate or revoke the development key when you finish the session.

Original data

Try it yourself

Open the OpenAI-compatible API tester →

Why point a playground at Plugsky

Playgrounds are the fastest way to iterate on prompts without writing a test harness. Pointing one at Plugsky lets you compare models on the same prompt in the same UI — something a single-vendor playground cannot do — and then hand the winning parameters to engineering as a concrete request body.

The Plugsky docs list the OpenAI Playground path as supported through a custom base URL field and tested regularly. The same pattern applies to other OpenAI-compatible playgrounds: find the provider settings, override the base URL, and supply a Plugsky key. Models are addressed by their ids — plugsky-micro, plugsky-lite, plugsky-pro and the rest of the 30+ catalogue.

Configuring the endpoint

The configuration is three values:

1. Base URL: https://api.plugsky.com/v1.
2. API key: a scoped sk-live-… key from the dashboard.
3. Model: the id you want to test, for example plugsky-pro.

Once the key is set, the playground usually calls GET /v1/models to populate its model dropdown. If the list loads, authentication and the base URL are correct. If it does not, check for a trailing slash on the URL, a missing /v1 segment, or whitespace pasted into the key field — these account for most setup failures.

Supported request controls map directly to the API: temperature, top_p, max_tokens, stop, streaming, JSON mode and tool definitions. What you set in the UI is exactly what the request sends.

What to test in the playground

Use the playground for prompt design and model selection, not for integration testing:

• Prompt iteration: adjust the system message and examples, then re-run with a fixed temperature.
• Model comparison: run identical prompts on plugsky-micro, plugsky-lite and plugsky-pro and note where quality actually changes.
• Parameter sensitivity: test low versus high temperature and tight versus loose max_tokens.
• JSON mode: confirm the model returns parseable JSON before you build a parser around it.
• Tool schemas: verify the model selects the right function for representative inputs.

When a configuration works, copy the exact parameter set into a test in your repository. A prompt that only exists in a browser cannot be regression-tested.

Key hygiene and fallbacks

Playground requests originate from the browser on your machine, so treat the key as exposed there. Create a dedicated development key with the narrowest scope that works, avoid admin, and revoke or rotate it when the session ends. Never paste a production key into a shared or hosted playground.

If your playground build does not expose a custom base URL field, you have two equivalent options: the playground inside the Plugsky dashboard, which is preconfigured for your account, or the OpenAI-compatible API tester, which validates keys and sends requests without any setup. Both cover the prompt-iteration loop; neither replaces running your own eval suite before shipping a prompt to production.

Honest comparison

Capability	Plugsky in a compatible playground	Vendor-locked playground	Plugsky dashboard playground
Model choice	30+ Plugsky models	Vendor models only	30+ Plugsky models
Base URL	Custom OpenAI-compatible endpoint	Vendor-hosted only	Built in
Auth	Your scoped sk-live-… key	Vendor account login	Dashboard session
Prompt iteration	Full parameters with export to code	Full parameters in-vendor	Full parameters
Model comparison	Switch models in one UI	Single vendor	Switch models in one UI
Setup time	Minutes	Minutes	No setup

Frequently asked questions

Does the OpenAI Playground support custom base URLs?

Builds that expose a custom provider or base URL field can point at https://api.plugsky.com/v1. The Plugsky docs list this path as tested. If your build lacks the field, use the dashboard playground or the API tester.

Which model should I start with?

plugsky-lite is a good general default, plugsky-micro is cheapest for simple tasks, and plugsky-pro is the strongest for complex prompts.

How do I authenticate in a playground?

Paste a scoped sk-live-… key into the API key field. The playground sends it as Authorization: Bearer on each request.

Is it safe to use a production key?

No. Browser-originated requests expose the key to your machine and any extension with access. Use a narrow development key and revoke it afterwards.

Can I test tools and JSON mode?

Yes, if the playground exposes those controls. Otherwise test them with the API tester or a short script against /v1/chat/completions.

Why does the model list fail to load?

Usually a malformed base URL or key. Confirm the URL ends with /v1, remove any trailing whitespace from the key, and verify the key has read access.

Can I save playground settings as code?

Most playgrounds show the equivalent request body or code snippet. Copy that into a repository test so the prompt is versioned.

What is not available in Plugsky yet?

Image generation, audio, moderation, files, batch, fine-tuning, assistants and the responses endpoint are coming soon; chat, streaming, tools, JSON mode and embeddings are live.