Real-time network monitor + outbound firewall — see & block every connection. Privacy-first, local-first, Rust fleet agent, SIEM exports, 8 languages. / Moniteur réseau temps réel + pare-feu sortant, agent de flotte Rust, exports SIEM. 100 % local.
Snitch — Enterprise integrations / Intégrations entreprise
EN: Step-by-step recipes for wiring Snitch into a SOC stack — Wazuh, Splunk, Graylog/Loki, Ansible, Docker/Kubernetes — all free and self-hosted, matching Snitch's local-first posture. Snitch emits no telemetry of its own; the only outbound connections are the ones you configure below.
FR : Recettes pas-à-pas pour brancher Snitch dans une stack SOC — Wazuh, Splunk, Graylog/Loki, Ansible, Docker/Kubernetes — gratuites et auto-hébergées, dans l'esprit local-first de Snitch. Snitch n'émet aucune télémétrie propre ; les seules connexions sortantes sont celles configurées ci-dessous.
1. SIEM output / Sortie SIEM
Snitch ships three native formats for alert forwarding:
| Setting / Réglage | Values / Valeurs | Effect / Effet |
|---|---|---|
syslog_addr |
host:514 or tcp://host:514 |
Enables syslog forwarding (UDP default, TCP with the tcp:// prefix) |
syslog_format |
rfc3164 (default) |
Classic BSD syslog — readable by every collector |
cef |
`CEF:0 | |
webhook_url + webhook_enabled |
URL | Generic JSON POST (or Discord/Slack/Telegram auto-detected payloads) |
GET /forensics/export?format=… |
json, csv, stix, ecs |
One-shot forensic bundles — ecs emits newline-delimited Elastic Common Schema |
Set them in the UI (Settings → Security) or in snitch.toml:
syslog_addr = "siem.example.com:514"
syslog_format = "cef"
2. Wazuh
Enable a syslog listener on the Wazuh manager (
/var/ossec/etc/ossec.conf):<remote> <connection>syslog</connection> <port>514</port> <protocol>udp</protocol> <allowed-ips>your-agent-subnet</allowed-ips> </remote>Point each Snitch agent at it — Ansible:
snitch_syslog_addr: "wazuh.example.com:514" snitch_syslog_format: "cef"Wazuh's CEF decoder picks the
CEF:0|Snitch|…messages up; alertseveritymaps to syslog PRI (critical→2, warning→4, info→6).
3. Splunk
- Settings → Data Inputs → UDP → new input on 514, sourcetype
syslog(or a TCP input fortcp://…senders). syslog_format = "cef"gives yousrc,dst,dpt,cs1Labelfields for free; for dashboards, pollGET /forensics/export?format=ecson a schedule through a scripted input and let the JSON auto-extract.
4. Graylog / Loki
- Graylog: System → Inputs → Syslog UDP, bind 514. CEF extension fields land as message fields.
- Loki/Grafana: point
syslog_addrat apromtailsyslog target or scrape/forensics/export?format=ecswith a cron →lokipush job.
5. Declarative configuration (Security-as-Code)
The backend applies one file at startup — $SNITCH_CONFIG, else
snitch.toml/snitch.json in the data dir. Every key it declares
overrides the UI/DB value, so the file is the fleet source of truth:
syslog_addr = "siem.example.com:514"
syslog_format = "cef"
kill_switch_enabled = true
ask_enabled = false # headless servers
gossip_peers = ["sec-01:8000"]
ports = [80, 443]
See packaging/snitch.toml.example for the full vocabulary.
6. Ansible / Puppet / Salt / GPO
Ansible — ready-made role in
deploy/ansible/:ansible-playbook -i inventory deploy/ansible/site.ymlIt installs
/opt/snitch, the venv,/etc/snitch/snitch.tomland the hardened systemd unit, then enablessnitch.service.Puppet / SaltStack — community skeletons in
deploy/puppet/anddeploy/salt/(labelhelp wanted): the same three primitives — copy the backend tree, rendersnitch.toml, managesnitch.service.Windows GPO —
deploy/gpo/Install-Snitch.ps1performs the silent install (ProgramData config, venv, SYSTEM scheduled task on 127.0.0.1) for GPO startup scripts or Intune win32 packaging. Ask-on-connect on Windows additionally needs the signed WFP callout — seedocs/SIGNING.md.
7. Docker & Kubernetes
Single host:
docker compose up—network_mode: host+cap_add: NET_RAW/NET_ADMINalready configured indocker-compose.yml.Fleet console:
deploy/helm/snitch—mode: collectorruns a ClusterIP Deployment that merges every configured agent's audit intoGET /remote/fleet;mode: agentrenders a host-network DaemonSet withNET_RAW/NET_ADMIN.helm install snitch ./deploy/helm/snitch \ --set token=$(openssl rand -hex 32) \ --set config.gossip_peers='["agent-1:8000","agent-2:8000"]'
8. Sigma rules & YARA-L
POST /alerts/rules/sigma accepts a Sigma detection YAML and compiles the
supported subset (field→value selections, |contains/|startswith/
|endswith/|re/|cidr/|gt/|lt/|all, 1 of/all of conditions)
into Snitch custom rules — alert-only, skipped reasons returned for
anything outside the subset. Fields map to what a packet sensor sees
(DestinationIp→remote_ip, Image→process_name, Domain→host, …).
YARA-L for Google SecOps/Chronicle: GET /alerts/rules/yaral exports all
custom rules (including Sigma-imported ones) as a .yar file — the
Sigma→YARA-L path is covered transitively.
9. Remote agents (snitch-agent/1)
agent/ ships a zero-dependency Rust daemon (~1 MB single binary) that
watches the host socket table (/proc/net/*, Linux) and pushes newly
observed outbound flows to POST /ingest/events. It observes sockets —
no payloads/SNI (that's the local capture's job); pid/process
attribution works on Linux via the /proc/*/fd inode walk and reports
0 elsewhere until the Windows/macOS backends land. Transport: HTTP,
same trust-domain model as gossip_peers. A snitch-agent.service
unit lives in deploy/systemd/; CI compiles and tests the binary on
Linux, macOS and Windows on every agent/** push.
Per-agent enrollment
Each deployed agent should carry its OWN credential — a stolen token can only impersonate that one agent, and revocation is per-name:
# EN: mint on the collector (master token), returned once
curl -X POST http://collector:8000/ingest/enroll \
-H "X-Snitch-Token: <master>" -d '{"name": "edge-01"}'
# → {"token": "sagt_…"} — put it in the agent's SNITCH_TOKEN
curl -X POST http://collector:8000/ingest/revoke \
-H "X-Snitch-Token: <master>" -d '{"name": "edge-01"}'
curl http://collector:8000/ingest/agents -H "X-Snitch-Token: <master>"
The ingest endpoint accepts X-Snitch-Agent: sagt_… (enrolled name is
authoritative — the body's host field can't override it) or the
master X-Snitch-Token for local/small deployments. Only token hashes
are stored, so a leaked settings dump can't replay credentials.
10. Threat-intel feeds (free)
POST /firewall/blocklist imports both shapes:
kind: "domain"— hosts/domain lists (StevenBlack, EasyList…) →domainfirewall rules;kind: "ip"— IP feeds (Blocklist.de, URLhaus/Feodo exports) →iprules; private/reserved/loopback addresses are never imported.
path: reads a local file (fully offline); url: is an explicit,
audited opt-in fetch (remote_fetch: true in the response). Cap at
max_entries ≤ 5000 — giant feeds belong in a resolver.
Opt-in API providers (AbuseIPDB / OTX / VirusTotal) stay available via
ti_provider + ti_api_key — nothing calls out until you set a key.
Scheduled refresh (ti_feeds)
A background worker re-imports feeds on their own clock — no restart, hot firewall reload. Configure it declaratively:
ti_feeds = [
{ url = "https://lists.blocklist.de/lists/all.txt",
kind = "ip", action = "block", hours = 24 },
{ path = "/etc/snitch/corp-domains.txt",
kind = "domain", action = "sinkhole", hours = 6 },
]
Each entry runs through the same importer as POST /firewall/blocklist — dedupe against existing rules, path: stays
fully offline, url: is the same audited opt-in. Last-fetch state
persists across restarts (ti_feeds_state); ≤64 feeds, ≤5000 rules
per refresh — giant feeds belong in a resolver, not here.
11. Headless / daemon mode
The GUI is optional: the backend is the whole product. Run it bare
(uvicorn api.main:app), under the shipped systemd unit, or in the
container — every feature above works without the frontend.
SNITCH_DEMO=1 still gives a no-root synthetic feed for training rooms
and CTFs.