Key facts
| Vision input | Image understanding runs on the OpenAI-compatible chat completions endpoint with supported multimodal models |
| Example model | plugsky-gemma-4 handles text plus image inputs such as screenshots, forms and product photos |
| Model access | 30+ models from efficient to frontier tiers behind one key |
| Dedicated media endpoints | Image generation and editing endpoints are labelled coming soon |
| Deployment | Hosted, VPC, on-prem and air-gapped options on Enterprise |
| Data handling | Data encrypted at rest and in transit; API data is not used to train models |
| Audit | Per-request logs with model, tokens, latency and user; SIEM export; retention up to 7 years |
| Free plan | plugsky-micro and plugsky-lite, no card; 14-day full-access trial |
TL;DR
- Classify image data flows before choosing a model or deployment.
- Split high-volume extraction from lower-volume reasoning across model tiers.
- Preprocess at the edge: resize, strip metadata and redact where required.
- Align residency and key custody with the sensitivity of each flow.
- Image generation is coming soon — plan a fallback before committing.
How it works, step by step
- Classify image flows by data sensitivity and allowed processing boundary.
- Separate extraction tasks from reasoning tasks and assign a model tier to each.
- Define the preprocessing contract: resizing, metadata stripping and redaction.
- Specify the output schema and the confidence threshold for human review.
- Map each flow to a deployment plane and key-custody requirement.
- Design audit fields, SIEM export and retention for image requests.
- Standardise the contract, then extend to further flows team by team.
Original data
Try it yourself
Open the model availability tracker →
Why enterprise architects need image understanding
Images complicate architecture because they carry more than text: faces, signatures, addresses, whiteboards, screenshots of internal systems. Before choosing a model, classify the image flows. Which images contain personal data, which are regulated, which may leave the tenancy at all? The answer determines deployment and, sometimes, whether the use case is viable as designed.
Then separate extraction from reasoning. Extraction is high-volume and forgiving — read the fields. Reasoning is lower-volume and needs a stronger model — interpret the document. Two tiers with different models, prompts and evaluation sets keep cost and quality manageable.
How vision works on the Plugsky API
The vision path is a chat completion with an image in the message content. Send the image alongside the prompt using the OpenAI-compatible request shape you already use for text, and the model returns a normal completion that you can stream, constrain with JSON mode or feed into function calling. plugsky-gemma-4 is the workhorse for high-volume visual triage — screenshots, forms, product photos and document images — while frontier models handle harder reasoning over the same input. For architects, the pipeline has three boundaries to design: preprocessing, the processing plane and the output contract.
The controls enterprise architects should configure first
A reference topology is straightforward: preprocess at the edge (resize, strip metadata, optionally redact), send the image through the OpenAI-compatible chat completions call with a strict output schema, validate server-side, and route low-confidence results to human review. Keep residency aligned to the corpus — region-locked planes for hosted work, VPC, on-prem or air-gapped deployments for the strictest flows — and keep keys with BYOK through your KMS or an on-prem HSM.
Audit is part of the design: per-request logs with model, tokens, latency and user, exported to your SIEM with retention up to 7 years. Note the boundary honestly — image generation endpoints are labelled coming soon, so do not put them on a roadmap without a fallback plan.
A practical pilot path
Phase adoption: one document flow, one team, one evaluation set, then standardise the preprocessing and validation contract before adding more. Prototype on the free plan and use Enterprise for SSO, private deployment and contractual terms. See the live pricing page for current plans.
Honest comparison
| Capability | Plugsky | Vision-only API vendor | Building in-house |
|---|---|---|---|
| Image understanding | Live on supported multimodal chat models | Usually the core product | You host and serve vision models |
| Image generation | Labelled coming soon | Often available today | Custom deployment work |
| API shape | OpenAI-compatible chat completions | Vendor-specific SDK | You standardise it yourself |
| Model choice | 30+ models behind one key | Narrower catalogue | One model per integration |
| Deployment | Hosted, VPC, on-prem and air-gapped | Usually hosted only | Your infrastructure only |
| Controls | Region pinning, BYOK, per-request audit logs, SIEM export | Varies by vendor | You build the pipeline |
Frequently asked questions
Where are images processed?
That follows your deployment choice: a region-locked cloud plane, your VPC, on-prem or an air-gapped environment. The API shape is identical, so the flow can move as classification demands.
Can we rely on image generation in our roadmap?
Not today — image generation and editing endpoints are labelled coming soon. Plan around image understanding now and keep a documented fallback for generation use cases.
Do we have to change our application code?
No. Image understanding uses the same OpenAI-compatible chat completions endpoint, so your SDK, auth and error handling carry over; the message content simply includes an image.
Is there a free plan?
Yes — plugsky-micro and plugsky-lite with 2 API keys and no credit card. A 14-day full-access trial is also available.
How is pricing structured?
Self-serve plans are flat monthly with unlimited fair-use usage and no per-token charges or overage fees. See the live pricing page for current plans.
Is image generation available?
Not yet — image generation and editing endpoints are labelled coming soon. Image understanding is live today on supported multimodal chat models.