The enterprise platform for AI agents · now in early access
Build your agents once.
Keep them current for good.
SparQ is where developers and businesses build, host, deliver and maintain capable AI agents and applications, with sovereignty built in, frontier models a policy away, and model upgrades that never need an application release.
- Swap models without a release
- Residency enforced on every call
- A credential on every action
- 1
- contract between your application and every model it will ever use
- 6
- levels of feature control, from platform down to a single conversation
- 30 s
- to refuse a revoked key everywhere. Measured at about ten.
- 0
- credentials in model context, run state, checkpoints or traces
Why it exists
The first AI feature is easy. The tenth customer, the first audit and the next model are not.
Every team that ships agents meets the same six walls. Most meet them one at a time, in production.
- 01Applications are hard-wired to one modelYour code names a model and a prompt. When a better or cheaper model ships, or the old one is retired, you change code, retest and redeploy. Every quarter.
- 02Every team rebuilds the same plumbingStreaming, retries, approvals, provider keys, budgets and logging get rebuilt in each application. Each copy holds its own keys and its own policy, so nothing is controlled in one place.
- 03Agents act with too much authorityA prompt is not a security boundary. Without one enforcement point, a tool runs with whatever privileges the agent happens to hold, for whichever customer happens to be asking.
- 04Spend you cannot attribute or stopToken costs land on one invoice. Nobody can say which customer, feature or credential spent what, and a looping agent keeps spending until someone notices.
- 05Residency and AI clauses in every contractCustomers now ask where their data is processed and which models may see it. That has to be enforced and provable, not configured by hand in each agent.
- 06Changes ship without a safety gateChanging an agent's instructions or tools usually means a code release. Or worse, it means editing a prompt in production with no evaluation and no second pair of eyes.
“Applications integrate with the Project, never with the model.”
A Project is your application's contract with SparQ: its agents, tools, knowledge, budgets and environments. Behind that contract, models, instructions and policy can be swapped, versioned and rolled back, and your application never notices, except that it gets better.
The life of an agent
Build, host, deliver, maintain, stay current.
Five chapters every agent goes through. SparQ is the one place all five happen, under the same rules.
- IBuildDescribe the agent. Don't hard-code it.Read more
- IIHostRun it on a platform built for many customers.Read more
- IIIDeliverPut it in front of people, three ways.Read more
- IVMaintainKnow what every agent did, and what it cost.Read more
- VStay currentTake the better model without touching the app.Read more
Security
Secure by construction, not by configuration.
Eight invariants hold for every workspace, every agent and every call. They are compiled into the platform; no setting turns them off.
- 01Authentication, and the credential recorded, on every action
- 02Isolation of workspace, project and client in data, cache, traces and budgets
- 03The tool gateway on every tool call, with default deny
- 04Approval for writes, and two people for destructive actions in production
- 05Hard limits on every run
- 06Audit writes that fail closed
- 07No credentials in model context, run state or traces
- 08Prompt and response content off by default in telemetry
Row-level security that no role bypasses
Early accessEvery table enforces row-level security, global tables included, and the build fails if a new table does not. No SparQ database role can bypass it, and a query without a tenant scope matches nothing.
Keys that are never stored
Early accessAPI keys are shown once. SparQ stores an HMAC with a pepper that never enters the database. A revoked key is refused within 30 seconds; measured at about 10.
Your identity provider, verified properly
Early accessUser tokens must be RS256 or ES256 with issuer, audience, expiry and subject checked. Symmetric algorithms are refused. Users get pseudonymous ids.
Fail closed
Early accessAuthorization, model routing and audit writes deny when they are unsure. A rate limiter that cannot be reached refuses the request instead of waving it through.
Credentials stay out of the model
Early accessNo credential ever enters model context, run state, checkpoints or traces. Provider keys are referenced by name in the registry; inline keys are refused.
Access scoped per environment
A key for development is refused by production. Production access is never implied by an older grant, and elevation is just-in-time with a second approver.
Two people for what matters
Publishing to production needs a second admin; the author cannot approve their own change. Destructive actions in production always need a person.
Egress you control
Tool calls run under an egress allowlist with SSRF and DNS-rebinding defences. Private and cloud metadata addresses are denied.
Pricing
Start with one application. Grow to sovereign.
Plans set seats, projects, environments and where SparQ runs. Model usage is metered from the same ledger you can read through the API.
Free
$0
forever
Try SparQ with one application.
Starter
$99
per month, from
A first application in production.
Pro
$999
per month
Teams shipping several AI features.
Business
$3,500
per month
Regulated workloads and larger customer bases.
Enterprise
Custom
annual agreement
Sovereign deployment and contractual terms.
Indicative pricing during early access.
Give your next AI feature a contract that outlives the model.
Tell us what you are building. We will show you SparQ on your use case and your residency rules.
Early access: Built and tested. Available to early-access workspaces.