Overview

The objects you work with in Sparq, and how they fit together.

The core idea

Applications integrate with the Project, never with the model. Each application gets a Project with a stable contract. Behind it, Sparq can swap models, update instructions, add tools and re-index knowledge, and the application does not change.

Objects

ObjectWhat it is
WorkspaceThe top-level owner: your organisation. Holds projects, people and budgets.
ProjectOne application's contract with Sparq: its credentials, token issuers, agents, skills, tools, knowledge, model classes and budgets.
EnvironmentA stage of a project, such as dev, qa, preprod or prod. Environments have a promotion order; a protected environment needs a second approver to publish.
ReleaseAn immutable bundle of a project's configuration: models, instructions, agents, tools, guardrails, knowledge and the evaluation baseline. Promoted and rolled back as a whole.
Client (audience tenant)One of your own customers, inside a project. Data, memory, traces and budgets are isolated per client.
Compliance profileA client's rules: allowed regions, providers and model classes, data sensitivity, retention. Model routing obeys it and fails closed.
CredentialAnything that authenticates a caller: a project key, a token issuer binding, a delegated tool token. Every request, cost row and audit record names the credential that caused it.
Agent specData that defines an agent: its job, instructions, model class, tools, skills, limits and the actions that need approval.
RunOne execution of an agent. It streams events, may pause for a person, and ends in a typed terminal state.

Every client-scoped record carries the isolation key (workspace, project, client). Each project also has a reserved client of its own, for work that is not on behalf of a customer.

Two APIs, one contract

PathUsed by
Runtime API/v1/...Your application at run time: sessions, messages, runs, approvals, features
Admin API/v1/admin/...Your back office and the consoles: tenancy, credentials, compliance, models, changes, usage, audit

Both are described by one OpenAPI document and versioned together. See versioning.

What Sparq never does

  • It never runs your code. It reaches your systems only through tools you expose (MCP servers, or operations imported from your OpenAPI document).
  • It never reads your databases directly.
  • It never puts a credential into a model's context, a run's state or a trace.

On this page