Vigo
Tens of thousands of envoys on one server — you set the check-in cadence, down to one second.
Vigo is a distributed state enforcement engine: agents on your managed nodes (envoys) pull desired state from a central server over gRPC/mTLS, apply idempotent resource changes, and report results back — at a tight cadence, at fleet scale, on commodity hardware. When one server isn't enough, spanner federates write-equal peers and the ceiling goes away.
But scale is only half of it. Vigo is far more than configuration management — it bundles the platform an operator usually has to assemble from a half-dozen tools: enforcement and compliance scoring and GRC evidence and reverse remote access and a peer-to-peer content fabric and an AI assistant and HA — under one mTLS-signed control plane. Vigo uniquely pairs elite tight-cadence scale with that full bundle.
What Vigo does
State enforcement & convergence. Describe desired state as configcrates — reusable resource sets across 69 idempotent resource types (file, package, service, user, exec, container, vm, repositories, the user-supplied custom executor, …). Resources form a depends_on DAG with notify/subscribes triggers; when: conditionals and Go-template content: adapt per host. The agent checks before it acts, self-heals from external failures, and converges offline from its cached bundle when the server is unreachable. Convergence is classified on two orthogonal axes — failure (Converged/Degraded/Failed) × drift (None/Changed/Diverged) — plus reachability.
Multi-axis configuration. One YAML language ties together configcrates, roles, hostcrates (hostname-pattern → role/vars, first match wins), per-user usercrates, and per-environment overrides, with a clear var-precedence chain. The source of truth is files only: edit the stacks, vigocli config publish, the server validates and reloads.
Compliance & GRC. Per-framework coverage (HIPAA, SOC 2, PCI DSS, ISO 27001, and more): a control counts as covered when a configcrate claims it (provides: capabilities or bundle: framework cuts), a Connwaer active check escalates it to satisfied, or a waiver applies. Push per-control pass/fail evidence to Vanta, Drata, Secureframe, ServiceNow GRC, and others on a schedule; export cyber-insurance / audit PDFs. Convergence and compliance are never conflated.
Reverse remote access (Scrier). Admin SSH / RDP / VNC to any envoy with no inbound ports on the node — sessions reverse out through the agent's existing stream, land in the right OS user's authorized_keys transiently, and never expose lateral movement. RDP/VNC ride a guacd sidecar in the browser.
Peer-to-peer swarm. A shared mTLS substrate carries six content subsystems — filecast (admin file push), gitback (git over swarm), longdrawer (LAN plaintext sync), lockbox (encrypted sync), curator (artifact registry + Docker/S3 shims), poolq (ordered-log queue) — all rooted in the per-user puddle identity. Fleet-distributed binaries, container images, and agent self-updates move node-to-node instead of hammering the server.
Orchestration. Dispatch ad-hoc tasks and live queries across the fleet; chain them into workflows with conditional branching and abort; roll changes out in health-checked batches.
AI assistant. A built-in, strictly read-only assistant answers questions about your fleet through ~14 tools (traits, runs, compliance, risk, CVE, docs), streamed, with the docs corpus in-context.
High availability & DR. Run peer replicas with one-command failover; or partition the fleet across spanner peers. SQLite with optional Litestream WAL replication to S3 keeps state durable.
Security by construction. Every byte is mTLS; every agent request is ed25519-signed against the stored pubkey; one-time enrollment tokens are hashed at rest; RBAC spans admin / viewer / compliance; a SHA-256 hash chain makes the audit log tamper-evident; startup hardening fail-closes on a bad binary signature; secrets resolve from a pluggable backend and never appear in config, logs, DB, wire, or run results.
Observability & integrations. Dashboards for convergence, compliance, risk (0–100 scoring), agent health, CVE impact, attack surface, and inventory — every number drilling down to its evidence. Prometheus on /metrics; outbound webhooks to Slack, PagerDuty, ServiceNow, ConnectWise, Splunk, Datadog, Loki, Opsgenie, Sentry, NetBox, and more.
The agent. A single Rust binary for Linux, macOS, and the BSDs / illumos (Windows supported; new subsystems ship Linux + macOS first). 69 resource types (including custom executors over JSON), 50 trait collectors, libvirt VM management, first-class containers, and signed self-update.
Start here
| Just want it running? | Quickstart — server up, first envoy enrolled, first configcrate applied, in ten minutes of copy-paste |
| New to Vigo? | Tutorials — the same path, walked step by step: stand up the server, enroll an envoy, ship your first configcrate, run your first compliance report |
| Need to do a thing? | How-to guides — configcrates, swarm, auth, monitoring, compliance, backup/DR, migrate from another CM |
| Looking something up? | Reference — vigocli, agent resources, agent traits, REST API, server.yaml, the configcrate language |
| Want the mental model? | Concepts — architecture, convergence vs compliance, the swarm, puddle identity, the security model |
| What's a term mean? | Glossary |
Confidential — Alexander4, LLC. Not for redistribution. See legal/license.md.