Key facts
| Tenancy model | Separate workspace and scoped keys per customer tenant |
| Shared codebase | One OpenAI-compatible integration across all tenants |
| Templates | Reusable agent, prompt and tool definitions per service line |
| Usage reporting | Per-key usage analytics for chargeback and client reporting |
| Pricing model | Flat monthly plans with unlimited fair-use usage on self-serve; see the live pricing page |
| Residency | Region-locked planes: EU (Frankfurt), GCC (UAE), APAC (Singapore), US (Virginia) |
| Deployment | Cloud, VPC, on-prem and air-gapped options for regulated clients |
| Endpoint status | Chat completions and function calling are live; assistants and responses endpoints are coming soon |
TL;DR
- One workspace and key set per client keeps tenant data and reporting separated.
- Build agent templates once and instantiate them per tenant to cut onboarding time.
- Flat monthly plans make managed-service margins predictable.
- Usage analytics per key turn AI consumption into a reportable, billable line.
- Offer VPC or on-prem deployment for clients who cannot use shared cloud.
How it works, step by step
- Productise two or three agent services — ticket triage, document summarisation or knowledge search — with fixed inputs and outputs.
- Create a workspace and scoped key set per client tenant, named and tagged for reporting.
- Build each service as a template: prompts, tool schemas and evaluation sets versioned once and deployed per tenant.
- Wrap delivery in your own portal so clients see your brand, not the upstream platform.
- Report usage per tenant from key-level analytics, and map it to your service tiers.
- Add a per-tenant audit view so clients can see what their agents did and who approved exceptions.
- For regulated clients, move the deployment to a VPC, on-prem or air-gapped tier while keeping the same integration.
Try it yourself
Open the AI agent cost calculator →
A tenancy model that scales
MSP economics depend on standardisation. Keep one codebase and one integration against the OpenAI-compatible endpoint, then separate tenants at the credential and data layers: a workspace and scoped key set per client, distinct vector indexes, and per-tenant audit streams. That gives you isolation you can describe in a security questionnaire without maintaining per-client forks.
Templates do the rest. A ticket-triage agent, a summarisation agent and a knowledge-search agent can each be defined once — prompts, tools, evaluation set — and instantiated for every tenant with client-specific data and escalation rules.
Packaging and margin
Flat monthly plans with unlimited fair-use usage make managed AI services easier to price than per-token resale, because your delivery cost does not swing with every conversation. See the live pricing page for current plans rather than working from stale numbers.
- Charge by outcome or seat: price the service, not the tokens.
- Watch heavy tenants: key-level usage analytics show which clients are outliers before they erode margin.
- Bundle controls: audit views and approval workflows justify a premium tier.
Delivery for regulated clients
MSPs increasingly serve clients who cannot use shared cloud infrastructure. The same integration runs in a VPC, on-prem or air-gapped deployment, with open-weight models where required, so you can offer a sovereign managed service without rebuilding the agent layer. Region-locked planes cover most residency requirements; perimeter deployments cover the rest.
Operationally, standardise on runbooks: onboarding checklist, tenant provisioning script, evaluation gate and offboarding purge. That discipline is what lets a small team manage dozens of agent tenants while keeping support and security review tractable.
Honest comparison
| MSP concern | Plugsky | Typical platform | Building in-house |
|---|---|---|---|
| Tenant isolation | Workspace and scoped keys per client | Varies; often account-level | You design tenancy |
| Delivery cost model | Flat monthly with fair-use usage | Often per-token | GPU and ops cost |
| White-label | Your portal and brand over one API integration | Vendor branding common | Entirely your product |
| Usage reporting | Per-key analytics for chargeback | Limited exports | Your metering |
| Sovereign delivery | VPC, on-prem and air-gapped tiers | Rarely supported | Possible, high effort |
| Model choice | 30+ models behind one endpoint | Vendor catalogue | You serve each model |
Frequently asked questions
How do we isolate customer data?
Provision a workspace and scoped key set per client, keep vector indexes inside the same boundary, and export per-tenant audit events. Isolation becomes a configuration step, not a code fork.
Can we put our own brand on it?
Yes. Plugsky is delivered through your portal and your contracts; the upstream platform is an implementation detail. White-label AI API guidance covers the packaging options.
How should we price managed agents?
Price the service, not the tokens. Flat monthly plans make delivery costs predictable; use key-level usage analytics to spot heavy tenants and adjust service tiers. See the live pricing page for plan specifics.
Do we need per-client deployments?
Usually not. Most clients are served from region-locked planes. Move a tenant to VPC, on-prem or air-gapped only when their regulation or contract demands it.
What reports can we give clients?
Per-tenant usage summaries from key analytics plus audit exports showing agent runs, tool calls, exceptions and approvals. Both are useful in quarterly service reviews.
How do we onboard a new tenant quickly?
Keep versioned service templates and a provisioning script. A new tenant is a workspace, keys, a vector index and client-specific escalation rules — hours, not weeks.
Can we start without upfront cost?
Yes. The free plan includes plugsky-micro and plugsky-lite with no card, and a 14-day full-access trial exists for a heavier pilot. Plan details are on the live pricing page.