osFoundry / Platform
Model playground
Test any model in a sandbox prompt → response interface with full token / cost accounting.
What it is
Model playground is the sandbox inside osFoundry where you type a prompt, pick a model, and see the response along with exactly what it cost. It is the fastest way to answer the questions that come up before you commit a model to a chat, an agent, or an automated loop: Does this model follow my instructions? How does its output compare to another model on the same prompt? How many tokens did that answer actually consume, and what did it cost me? Because osFoundry is bring-your-own-key, the cost shown is the provider's real price against your own account — there is no reseller markup sitting between you and the model.
Inside osFoundry the playground draws on the same model connections the rest of your workspace uses, so any model you have wired up is testable here: a local model running on your own machine, a cloud model reached with your own API key, or a model you self-host. You send a single prompt, get a single response, and the interface reports the token accounting for that exchange — prompt tokens in, completion tokens out — and the cost derived from them. That makes the playground the place to shake out a prompt and read the true per-request economics before that model does real work at scale, where a small per-call difference multiplies across every run.
Being a desktop, local-first workspace matters for what you can test here. When you point the playground at a model running on your own device, the prompt and the response never leave your machine, so you can iterate on sensitive or proprietary prompts without shipping them to a third party. The same prompt → response → cost loop works whether the model is local, cloud, or self-hosted, which is what lets you make an honest side-by-side call about quality and price across very different backends from one interface.
Key capabilities
- Send a prompt to any connected model — local, cloud, or self-hosted — and read the response in a single sandbox interface
- See full token accounting for every exchange: prompt tokens in, completion tokens out
- Get real per-request cost from your own BYOK provider pricing, with no reseller markup added
- Try prompts against a local model on your own device so sensitive input never leaves your machine
- Shake out prompt wording and measure true cost before wiring a model into a chat, agent, or loop
Use cases
- Developer choosing a model: run the same prompt against a local model and a cloud model, then compare output quality against token cost before committing
- Prompt engineer: iterate on wording in the sandbox and watch how each revision changes token count and cost
- Privacy-conscious analyst: test prompts with confidential data against a locally hosted model so nothing is sent to an external API
Frequently asked questions
Which models can I test in the playground?
Any model you have connected in osFoundry — a model running locally on your machine, a cloud model reached with your own API key, or a model you self-host. The playground uses the same connections as the rest of the workspace, so if you can use a model elsewhere in osFoundry you can test it here.
How is the cost calculated, and is there a markup?
Cost is derived from the token accounting of your request against your provider's own pricing. osFoundry is bring-your-own-key, so you pay the provider directly and there is no reseller markup added on top — the number you see reflects the real price of that call on your account.
Can I test prompts privately without sending data to a cloud provider?
Yes. osFoundry is a local-first desktop workspace. When you point the playground at a model running on your own device, the prompt and the response stay on your machine and are not sent to any external service, which is useful for iterating on sensitive or proprietary prompts.