Key facts
| API compatibility | OpenAI-compatible chat, embeddings and function calling |
| Use cases | In-product assistants, document Q&A, onboarding help, workflow automation |
| Tenant isolation | Scoped API keys and projects per tenant or environment |
| Data residency | Region selection for the markets your customers require |
| Deployment | Plugsky cloud, your VPC, on-prem and air-gapped options |
| Models | 30+ models from free aliases to frontier reasoning |
| Pricing model | Flat monthly self-serve plans with unlimited fair-use usage |
| Free tier | plugsky-micro and plugsky-lite on the free plan, no card required |
TL;DR
- Ship AI features with the SDK your team already uses.
- Scope keys per tenant so isolation is enforced outside prompt logic.
- Pick a processing region per customer commitment, not per feature.
- Route cheap requests to small models and reserve frontier models for hard ones.
- Prototype free, then benchmark on your own product data before rollout.
How it works, step by step
- Choose one feature that fits the product surface, such as document Q&A or a support assistant.
- Define the tenant boundary: key scope, retrieval index and retention per customer.
- Build on the OpenAI-compatible endpoint and keep prompts, tools and evals in version control.
- Add a routing layer so simple classification uses a small model and complex drafting uses a stronger one.
- Configure region selection and audit logging before the first customer pilot.
- Measure activation, deflection and support cost, then expand to the next feature.
Try it yourself
Open the AI workload router simulator →
AI features SaaS products ship first
The best first features reduce user effort using data the product already holds:
- In-product assistant: answer questions about the customer's own workspace with citations.
- Document Q&A: let users query contracts, tickets or records they uploaded.
- Onboarding help: explain setup steps and settings from your documentation.
- Workflow automation: draft, classify and route records using function calling.
All four work with the same compatible endpoint, so one integration covers several features.
Multi-tenant isolation and residency
Isolation should not depend on prompt instructions. Use scoped API keys per tenant or environment, separate retrieval indexes per workspace, and audit logs that record tenant, user and model for every request. Where customers require data to stay in a region, select the processing region at the integration level and document it in your own trust materials. For customers with stricter requirements, deploy a dedicated VPC or on-prem instance rather than sharing a public tenancy. Because the API is compatible, the same product code serves both models of delivery.
Cost control and model routing
AI features can fail commercially if every keystroke hits a frontier model. Route by task: small models for classification, extraction and short replies; mid-tier models for retrieval answers; frontier models for multi-step drafting or reasoning. Cache stable answers, cap retries and set per-tenant fair-use limits to protect the platform. Self-serve plans are flat monthly with fair-use usage, so product economics stay predictable while you tune routing. Freeze model versions for customer-facing features so quality does not shift without a release.
From prototype to customer-facing release
Prototype on plugsky-micro or plugsky-lite with synthetic or demo data, then use the 14-day full-access trial to benchmark frontier models on anonymised product data. Build an evaluation set before launch and re-run it on every prompt or model change. Track per-feature deflection, latency at peak and cost per active workspace. Roll out behind a flag to a subset of tenants, keep a kill switch, and expand only after the evaluation set is stable across model versions.
Honest comparison
| Capability | Plugsky | Typical per-token API | Building in-house |
|---|---|---|---|
| API compatibility | OpenAI-compatible chat, embeddings and tools | Usually compatible | Full rewrite |
| Tenant isolation | Scoped keys and separate projects or deployments | Shared key by default | You design tenancy |
| Data residency | Region selection per commitment | Often limited regions | You control fully |
| Pricing | Flat monthly self-serve, fair-use usage | Per-token, harder to forecast | GPU plus operations cost |
| Model choice | 30+ models behind one API | Varies by provider | You host every model |
Frequently asked questions
Can we keep our existing OpenAI SDK code?
Yes. Plugsky exposes an OpenAI-compatible API, so your SDK, retries and streaming handling stay the same.
Is there a free plan?
Yes. The free plan includes plugsky-micro and plugsky-lite with no credit card, which is enough to prototype a feature.
How do we isolate tenants?
Use scoped API keys per tenant or environment and separate retrieval indexes per workspace. Isolation is enforced by key scope, not by prompt text.
Can customers choose where data is processed?
Region selection is available, and a dedicated VPC, on-prem or air-gapped deployment is possible for stricter customer requirements.
How do we keep AI costs predictable?
Route tasks by difficulty, cache stable answers, set fair-use limits, and use flat monthly self-serve pricing. See the live pricing page for current plans.
Which endpoints are live today?
Chat, streaming, JSON mode, function calling, embeddings, RAG and agents are live. Audio, images, moderation, files, batch, fine-tuning, assistants and responses are coming soon.
Can we white-label the AI feature?
Yes. The API is invisible to your users, so AI features appear as part of your product. Check the docs for branding and resale specifics.