← Home

Votrix Managed Agents / VMA

Infrastructure for multi-tenant AI agents.

An API-first harness for multi-tenant AI agent products, with organization-scoped resources, sandboxed sessions, resumable events, and usage controls.

Created by Sonny Proto · AI Engineer

From an agent loop to a product API.

VMA exposes Agent work through reusable resources: Agents, Environments, Sessions, Accounts, Files, and Memory Stores. A product backend can define work, start a sandboxed Session, follow its events, and collect the resulting files.

My engineering focus is the system around the model: API boundaries, session lifecycle, durable state, tool execution, and usage controls. VMA brings that work together behind one public API contract.

Organization-scoped APIs

An API key selects the Organization it can access. Resource lookups stay within that boundary; missing and out-of-scope resources return the same not-found response.

Read the public contract

Sandboxed sessions and versioned agents

A Session connects an Agent version, an Environment, an Account, and attached Files or Memory Stores. It keeps its selected resources while follow-up work continues.

Read the public contract

Resumable event streams

Each Session has an ordered event history. Clients can follow progress over SSE, save the latest sequence, and resume after disconnecting.

Read the public contract

Usage attribution and spending controls

Accounts group Agent usage and spending. Each Session keeps one Account, so applications can attribute work to a customer, team, or workflow.

Read the public contract

Keep the integration boundaries explicit.

A product still owns its end-user authentication, authorization, resource mapping, and customer-facing billing. Accounts attribute usage and spending; they do not create an independent authorization boundary inside an Organization.

These capabilities describe VMA's public API contract. Performance, isolation requirements, and operational behavior should be evaluated for the application being built.