Agent Governance for Regulated Engineering Teams

Teams in regulated environments ask a different first question about agent automation: can it run where we are, and can we prove what it did. Four properties answer it, and Modulo is built around them.

Placement

Self-hosted is the placement model. Modulo runs on Docker Compose, a native Linux install that bundles its database and queue, or a host you manage. A deployment can sit on a network with no route to the public internet. There is no hosted edition, so no part of the system requires a vendor's infrastructure.

Data location

Prompts, pipeline definitions, agent outputs, run records, and audit events are written to your Postgres database, encrypted credentials included, with FERNET_KEY held in your environment. With default settings and no connectors configured, Modulo makes no external network calls.

Control points

Approval gates pause a pipeline until a named human decides, and a human-only gate cannot be claimed or decided by an agent principal. Organization roles and team roles scope what each person can do. Daily spend limits cap model cost at the organization and team level.

Evidence

Every state-changing action appends an audit event, chained with the SHA-256 hash of the event before it, append-only at the application and database layers, verifiable on demand, and exportable. Each run also holds the prompt, the output, the schema result, and the gate decisions.

You keep the custody

Modulo runs in your environment. Your data, your model agreements, and your audit trail stay yours - Modulo supplies the orchestration layer, not the custody. The deployment and its data sit under your control, and the audit record is evidence your own controls produce.

By sector

• Government and defense: placement inside an isolated network, agent execution on your own engine.
• Financial services: approvals, the model record, and evidence for an audit or supervisory request.
• Healthcare: clinical and patient-facing systems, data placement, and the record of a change.
• Own your LLM spend: your provider contracts, per-run token accounting, and no vendor share of model spend.