Key facts
| Tool type | Free static compatibility audit for OpenAI-based codebases |
| Checks | SDK and language, endpoints, model names, streaming, tools, JSON mode |
| Labels | Drop-in, adapter required, or roadmap/not yet available |
| Live on Plugsky | Chat, streaming, JSON mode, function calling, embeddings, RAG, agents |
| Roadmap items | Audio, images, moderation, files, batch, fine-tuning, assistants, responses |
| Migration effort | One line for OpenAI-style chat calls: base URL plus model name |
| Free plan | 2 free AI models (plugsky-micro, plugsky-lite), no card required |
| Product status | Live |
TL;DR
- Audit endpoints and features before code: chat is easy, specialist endpoints are not always ready.
- Label every usage drop-in, adapter or roadmap, and plan around the gaps.
- Keep a rollback path: an OpenAI-compatible endpoint makes reversal a config change.
- Test on your real prompts and schemas, not a hello-world call.
- Migrate one workload at a time and canary before full cutover.
How it works, step by step
- Inventory your codebase: SDKs, base URLs, endpoints and model names in use.
- List the features you depend on, including streaming, JSON mode, function calling and embeddings.
- Run the OpenAI migration checker and review the per-item labels.
- For adapter items, write the smallest translation layer and pin it with tests.
- For roadmap items, keep the current provider and isolate those calls behind an interface.
- Run your eval set against the migrated workloads with the API tester.
- Canary one workload, monitor error rates, then expand and keep rollback ready.
Try it yourself
Open the OpenAI migration checker →
What the checker inspects
The audit starts from your code and configuration, not from marketing pages. It looks at the client SDK and language, the base URL, every endpoint you call, the model names you hard-code, and the features that change request or response shapes: streaming, JSON mode, function calling, embeddings and file handling. Each finding gets a label. Drop-in means the same call works after a base URL change. Adapter required means the concept exists but the payload or behaviour differs. Roadmap means the capability is not available yet and should stay on your current provider for now.
Planning around the gaps
The labels turn a migration into a schedule. Migrate drop-in chat and embedding workloads first, because they build confidence and prove your observability works. Wrap adapter items in a thin provider interface so the translation lives in one file rather than scattered call sites. Keep roadmap workloads — audio, images, moderation, files, batch and fine-tuning are not live on Plugsky today — on their existing provider behind the same interface, with a review date tied to the docs. This layered approach lets you move most volume without blocking on the last feature.
Testing and rollback
A migration plan is not done until rollback is boring. Run the same prompts through both providers with the API tester, compare outputs, token usage and finish reasons, and log both so regressions are visible. Cut over behind a feature flag with a percentage rollout. Because the endpoint is OpenAI-compatible, rollback is the same configuration change in reverse — no redeploy, no schema migration. Keep the old credentials active until the new path has handled a full traffic cycle without surprises.
Honest comparison
| Stack item | Drop-in on Plugsky | Adapter required | Roadmap (stay put for now) |
|---|---|---|---|
| Chat completions | Yes, OpenAI-compatible endpoint | No | Not applicable |
| Streaming | Yes | No | Not applicable |
| JSON mode | Yes | No | Not applicable |
| Function calling | Yes | Schema details may need review | Not applicable |
| Embeddings and RAG | Yes | No | Not applicable |
| Audio, images, files, batch, fine-tuning | No | No | Yes, keep current provider |
Frequently asked questions
What does the migration checker do?
It audits your OpenAI usage — SDKs, endpoints, model names and features — and labels each item as drop-in, adapter required or roadmap, so you can plan the migration workload by workload.
Is the checker safe to run?
It performs a static compatibility audit based on the stack details it asks for. Do not paste production secrets into any audit tool; rotate keys you no longer need.
Which endpoints can move today?
Chat completions, streaming, JSON mode, function calling, embeddings, RAG and agents are live. Audio, images, moderation, files, batch, fine-tuning, assistants and responses are roadmap items.
How much code changes for chat?
For OpenAI-style chat calls, the base URL and the model name. Your SDK code and request shape stay the same.
What if I use features that are not available yet?
Keep those calls on your current provider behind a provider interface, and set a review date tied to the Plugsky docs so you can move them when they go live.
How do I test the switch?
Run the same prompts through both providers with the API tester, compare outputs and token usage, then canary one workload behind a feature flag.
How do I roll back?
Change the base URL and model name back. Because the API is OpenAI-compatible, no code refactor is needed for chat workloads.
Is there a free way to start the migration?
Yes. The free plan includes 2 free AI models (plugsky-micro and plugsky-lite) with no card required.