Sovereign agent action & governance control plane

Every agent action, journaled.
Every risky step, approved.

Vigiloz runs agentic security automation on a durable, replay-journaled engine: every API, LLM, and approval call is recorded before results flow, humans approve the steps that matter, and the whole run is auditable after the fact. Self-hosted in your cloud, on your models — Anthropic, OpenAI, Vertex, or fully local.

Request a demo → Why Vigiloz ↓

300+ SPEC-LOADED INTEGRATIONS · BYO-MODEL · SELF-HOSTED · MULTI-TENANT

EXECUTION wf_7f3a · phishing-triage REPLAY-SAFE

#0041atp.llm.classify(alert_9241)✓ journaled

#0042atp.api.edr.contain(host-1189)✓ committed

#0043atp.approval.request(tier-2)⏸ suspended

#0044resume(checkpoint #0043)▶ approved 42s

#0045atp.api.itsm.create(SEC-2114)✓ committed

JOURNAL · 47 calls · 0 replay gaps WORKSPACE soc-eu · credentials brokered at runtime

Autonomy without the black box.

THE PLATFORM — 01/03

A durable engine underneath

Workflows are plain TypeScript with real loops and conditionals, run in an isolated sandbox. Every call is journaled before results flow and replayed deterministically on resume — non-idempotent writes record intent first and fail closed.

Approvals woven into the run

An approval step suspends execution mid-run; a human decision resumes it from the exact checkpoint. Risky actions wait for the sign-off you configured — and Vigiloz learns from every decision made.

Your cloud, your models

Self-hosted, multi-tenant and MSSP-ready, with scope-based RBAC and immutable audit logs. Bring your own model — Anthropic, OpenAI, Vertex, or fully local with Ollama.

300+ integrations, loaded from their specs OPENAPI · GRAPHQL · SQL · MCP · POSTMAN · SDK — credentials brokered per workspace

Every action, accounted for.

HOW IT WORKS — 02/03

WRITE

Workflows are code — versionable and testable. Write them by hand, or describe the play and let Vigiloz's AI authoring draft workflows and agents for you.

export default async (atp) => …

JOURNAL

Every API, LLM, and approval call is recorded by execution and call index before results flow. Replays are deterministic; journal gaps fail closed.

journal(execution_id, call_index)

APPROVE

Approval steps suspend the run and wait for a human. The decision resumes execution from the exact checkpoint — minutes or days later.

approval.request → resume(ckpt)

AUDIT

Immutable audit logs and a replayable journal mean any run can be walked step by step, long after it finished — every action attributable.

replay(wf_7f3a) · audit trail

Governed autonomy, running in your cloud.

Durable execution, human approvals, and a full audit trail for agentic security operations — we'd rather show you a live run than tell you about it.

Request a demo →