Who DMZAgent is for.
DMZAgent is at the design-partner stage. There are no public customer logos yet, and this page would rather describe the three teams the platform is built around than invent testimonials. Each archetype below names the job, the work DMZAgent does on their behalf, and the reason a generic observability stack will not finish the job.
The compliance lead owning agent oversight in a regulated industry.
A bank, an insurer, a healthcare network, or a regulated platform now operates AI agents that take actions inside the business. The compliance lead is accountable for the record those agents produce. Their auditors are asking sharper questions every cycle, and the application logs the engineering team writes are not an answer.
A defensible record, ready to file.
- Signed, time-stamped events for every consequential decision an agent makes, with the evidence behind it linked.
- One-step audit packet export for any agent, any window, with a tamper-evident hash chain that auditors can verify independently.
- Risk-tag schema aligned to common frameworks (OWASP LLM, NIST AI RMF) with the option to import the team's existing taxonomy.
- Controlled-record workflow with two-party release, retention windows, and a ledger entry for every access.
- Auditor role with read-only access scoped to the records under review, separate from operational seats.
Application logs are not evidence.
- Most observability stacks are built for engineers debugging the agent, not for auditors reviewing it months later.
- DMZAgent treats the record as the product, not a side effect of tracing.
- The retention model is set by the customer, not by a vendor's default log retention.
- Sub-processor list, data residency, and a DPA are published. Procurement teams do not have to chase them.
- Security posture is laid out in plain language so the control matrix lands on the risk team's desk on day one.
The platform team running production agents at scale.
A platform team supports dozens of agents written by other teams across the organization. Latency dashboards are good enough for uptime, and bad at telling the platform owner that one of their agents has quietly drifted. When something breaks, the platform team is the one re-creating what happened from monitoring screenshots at 2 a.m.
A registry and a forecasting layer for the fleet.
- A current profile of every agent — capabilities, signature behaviors, known failure modes, risk tags, and the evidence behind each claim.
- Drift detection per agent and cohort, scored against the agent's historical record, surfaced as alerts to the team's existing paging system.
- Calibrated outcome forecasting so the platform team can answer "is this still working" with a number that has been Brier-scored, not a dashboard color.
- SDK in Python, TypeScript, and Go, plus a sidecar proxy when the runtime is closed and the agent code cannot be changed.
- Replay against historical traffic when a new agent version ships, so regressions are caught before the rollout reaches end users.
Built for fleets, not for one agent.
- Generic tracing tools record spans. DMZAgent records decisions — the action taken, the evidence behind it, the risks attached.
- Adding the SDK costs five lines around the agent loop. Most platform teams are in production within a week.
- The reasoning plane is swappable: bring your own model, or use the platform default. See the integration docs.
- The free tier is generous enough to instrument an agent end-to-end before procurement is involved.
- Release cadence is published; the platform tells the team what changed and why.
The internal-tools team adopting LLM workflows that need an audit trail.
An IT or operations group is bringing language-model workflows into the business — customer support triage, contract review, invoice handling, ticket routing. The legal team has asked who decided what, and when. The engineering team is small, the budget is bounded, and building a custom audit pipeline is not on the roadmap.
An audit trail without building one.
- A workspace in under a minute, with a generous free tier that covers the volume of an early internal rollout.
- A turnkey audit trail — every agent decision, the inputs, the evidence, and the outcome — readable by a non-engineer in the dashboard.
- Connector configuration via schema-form so non-engineers can plug DMZAgent into the team's existing tooling without writing code.
- Notification inbox so the team learns about drift, ingest failures, or policy violations in the dashboard, not by reading log files.
- Role-scoped access so the legal reviewer sees what they need to see and nothing else.
The simple version is the same as the regulated version.
- The same tamper-evident ledger that backs a regulated deployment backs the internal-tools workspace. There is no "audit-ready edition" sold separately.
- The dashboard is built for the non-engineer reviewer as well as the developer.
- Pricing is metered: the workspace stays inside the free tier until the volume justifies a paid plan. No seat-license land mines.
- When legal asks for a record, the team has one. When the workflow is retired, the record persists for the retention window the team chose.
- Security posture and the DPA answer the questions procurement raises before they are asked.
The design-partner program is open.
If one of the archetypes above is recognizable, the DMZAgent team would like to hear about the work. Design partners get direct access to the product team, priority on the roadmap, and a discount that holds through general availability.
The conversation starts with a thirty-minute call about the agents your team operates today, the records your auditors or your regulators are asking for, and where the current setup falls short. There is no obligation past that call.