🛡️ AutoGov

Control plane for shadow automation — find the unmanaged n8n instances leaking access to your production systems

Employees spin up self-hosted n8n in Docker on laptops and servers, wire in live credentials to 1C / CRM / databases / payment APIs, and run workflows without security's knowledge. Existing Shadow-IT tools see SaaS/OAuth — they do not see locally-deployed automations. AutoGov does.

find → document → protect → heal

Quick start · Usage guide · Architecture · Go-to-market · Русская версия

---

_{The dashboard (UI available in English and Russian): risk-scored findings with plain-language explanations, the “instance → credentials → target systems” access map, and sensor inventory.}

---

Why AutoGov

• 🔍 Sees what others miss. SaaS-discovery tools (Defender for Cloud Apps, Nudge, Auvik) find OAuth apps. EDRs inventory processes. Neither maps a shadow localhost:5678 n8n instance to the 1C and payment systems it can reach. AutoGov builds exactly that graph.
• 🧭 Access map, not just an inventory. For every unmanaged instance: which credentials → which production systems, with a numeric risk score, a Critical/High/Medium/Low category, and a human-readable “why” for the CISO report.
• 🔒 Privacy by design. The agent records the fact a secret exists — its name, type and target — but never reads or transmits the value (only a SHA-256 fingerprint). No field in the data model can hold a secret or PII.
• 🇰🇿 On-prem / in-country by default. In on-prem mode all external egress (including external LLMs) is refused — no cross-border transfer (Kazakhstan Law No. 94-V). Fully air-gappable: no external dependencies.
• 🧩 Platform, not a monolith. A universal core (agent + control plane + data model + event bus); M1 Discovery is the first vertical slice, M2–M4 plug in as modules without touching the core — proven by tests.
• 🔐 Secure by construction. Outbound-only agent (no listening ports), mTLS, read-only Docker socket, RBAC + audit log, and signed release artifacts the agent verifies at startup (fail-closed).

How detection works (multi-signal, low false positives)

The agent detects n8n by a combination of signals — never one — to keep false positives low:

Signal	What it inspects
🐳 Docker	n8nio/n8n images, port 5678, N8N_* env, linked postgres/redis (queue mode) — via read-only socket
⚙️ Processes	Node.js n8n signatures, Python runners/schedulers in /proc
🌐 Network	HTTP fingerprint: /rest + /api/v1/docs + /healthz uniquely identify n8n
📁 Filesystem	~/.n8n, docker-compose.yml with n8n, .env referencing N8N_ENCRYPTION_KEY
⏰ Schedulers	cron entries and systemd timers launching automation runners

Legitimate (IT-sanctioned) instances go on a whitelist so the product doesn't cry wolf on CI/CD containers — a first-release requirement, not a “later”.

🚀 Quick start (2 minutes)

Requires Docker with the Compose plugin.

git clone https://github.com/oleg-vdv/AutoGov.git
cd AutoGov/deploy
docker compose up --build

This brings up the control plane, an intentionally shadow n8n instance, and an agent. Open http://localhost:8443, log in with token dev-admin. Within ~30 s the agent discovers the n8n instance, builds the access map, and shows risk-scored findings.

Demo uses HTTP and dev tokens — local evaluation only. For a pilot, enable TLS/mTLS, replace tokens, and keep mode: onprem. See the usage guide.

Build from source

make build   # → bin/agent, bin/controlplane, bin/autogov-sign
make test

Pure Go standard library — no external dependencies (auditable agent, air-gapped installs).

Components

Path	Purpose
cmd/controlplane	Central server: ingest, API, UI, modules
cmd/agent	Host sensor (outbound-only), collectors + disk spool
cmd/autogov-sign	Release signing (ed25519) + integrity verification
internal/modules	Module Runtime + discovery (M1) and M2–M4 plugin stubs
internal/risk	Risk scoring with local (no-LLM) explanations
internal/agent/sanitize	Privacy guard: secret values never leave the host
internal/notify	Notifiers (syslog/CEF, webhook, email) + on-prem egress guard

Roadmap

• Stage 1 — M1 Discovery for n8n (MVP, pilot-ready)
• Stage 2 — generic automations (Make agents, scripts), Windows agent, more SIEM
• Stage 3 — M2 HoneyNodes (contract + stub already in place)
• Stage 4 — M3 Self-healing / M4 Autodoc (contracts + stubs already in place)

Business & positioning

AutoGov follows an open-core model: the agent and platform core are open (auditability removes the adoption barrier for a privileged sensor); paid modules M2–M4 are the up-sell. Full ICP, value-by-role, competitive moat and a 90-day validation plan are in docs/GO_TO_MARKET.md.

Contributing & security

See CONTRIBUTING.md and SECURITY.md. The privacy invariant is non-negotiable: a change that makes the agent transmit secret values will not be merged.

License

AGPL-3.0-or-later — see LICENSE.

• Scan your own estate, change the code, run it wherever you like — freely.
• Ship it inside a product or a service you sell — an MSSP offering, a security platform, a hosted scanner — and the AGPL obliges you to release the source of that product under the same terms.
• If that does not work for you, a commercial licence removes the obligation: see COMMERCIAL.md.

Releases up to and including the last Apache-2.0 tag stay under Apache-2.0 — nothing already published is taken back. The change applies from here onward, and the copyright is held by a single author.

_{Built to give security teams a control plane for the shadow-automation boom. Not affiliated with n8n GmbH — AutoGov inspects instances you already run, it does not host n8n.}