Gate first. Audit, recover, and execute second.
Novyx is organized around production action control. Memory, search, and context remain in the stack, but they support governance rather than define it.
Primary layer
Protected action gate
Agents request effects through Novyx before they touch production systems. The gate returns allow, deny, require approval, or require multi-approval.
Learn more →Blast-radius review
Novyx surfaces affected systems, records, users, jobs, and irreversible operations before execution.
Learn more →Approval workflows
Route production migrations, refunds, account changes, emails, CRM writes, and exports to the right reviewer.
Learn more →Operational layer
Everything after the gate is evidence and recovery.
Once an action is classified, Novyx follows it through decision, execution, audit, replay, and recovery. The result is a system of record for agent-caused production change.
Audit
Tamper-evident proof of each request, decision, approval, denial, execution result, and recovery event.
Recover
Rollback and compensation metadata captured before risky work runs, not after the incident starts.
Execute
Governed connectors let agents act through controlled paths instead of raw credentials.
Replay
Review how the agent moved from intent to tool call to outcome when something needs investigation.
Support layer
Memory is still useful. It is not the headline.
Context features help explain why an agent made a request, what it has done before, and how similar incidents were handled. They strengthen the gate rather than replace it.
Memory
Retain useful context, prior decisions, and run history as evidence for future reviews.
Knowledge graph
Map relationships between agents, tools, systems, records, and incidents.
Eval signals
Detect drift in action patterns and approval outcomes over time.
MCP support
Expose governance workflows to MCP-capable coding tools without making MCP the product story.
Feature map
Protect
Action gate, risk tiers, blast-radius review, approvals.
Operate
Connectors, execution controls, audit export, replay.
Support
Run context, incident history, knowledge graph, eval signals.