Introduction to Stackmint

Stackmint is a governed AI execution platform built for enterprises that need more than just model access. It provides the infrastructure to build, deploy, and control AI workflows across teams — with audit trails, budget enforcement, model governance, and human approval gates baked in at the platform level.

The problem it solves

Most AI tooling focuses on model access. Stackmint focuses on what happens after: who can run what workflow, on which model, with what budget, and with whose approval. These are not afterthoughts — they are the conditions under which AI can safely operate in regulated industries, enterprise procurement cycles, and cross-functional teams.

Without a governance layer, teams end up with:

• Shadow AI deployments with no visibility
• Uncontrolled model costs with no budget attribution
• No audit trail when a workflow produces incorrect output
• No mechanism to halt a misbehaving Capability in production

Stackmint solves all of these at the infrastructure level, not the application level.

Key primitives

Every Stackmint deployment is built from a small set of composable primitives:

• Buds — Typed, versioned execution units. Each Bud has a defined input schema, output schema, and implementation. Buds can call models, read databases, send messages, or invoke external APIs.
• Branches — Directed execution pipelines that chain Buds together. A Branch defines the expected inputs, the sequence of Bud calls, and the final output shape.
• Capabilities — Workspace-deployed instances of Branches, with governance policies applied via a Manifest. A Capability is what end-users interact with.
• Surfaces — The interface through which Capabilities are exposed: Chat, Client Cockpit, Partner Cockpit, or embedded iFrame.
• Workspaces — Isolated multi-tenant containers that group users, Capabilities, budgets, and governance configuration.

Governance model

Governance in Stackmint operates at multiple layers simultaneously:

• Model layer — Each Workspace maintains an allow list of approved models. No Bud can call a model not on this list. Model routing rules can further restrict which teams or Capabilities can access which models.
• Budget layer — Credits are allocated per team, per Branch, and per Capability. Executions that exceed their budget ceiling are terminated before the model is called.
• Approval layer — Capabilities can require human-in-the-loop checkpoints before sensitive Buds execute. Approval routing is policy-driven by role, department, or execution value.
• Audit layer — Every execution is logged with a cryptographically sealed trace. Logs are immutable and exportable.

What it is not

Stackmint is not a model provider, a chatbot builder, or a no-code workflow tool. It is infrastructure for teams that already know what they want to build with AI and need a foundation that will survive production scrutiny, compliance audits, and organizational scale.

Next steps

Continue with the Quickstart to set up your first Workspace and run a Capability in under 10 minutes, or read the Architecture Overview for a deeper look at how the platform is structured.