Open Source · by Laboratório Hacker

Your AI Agents
Can Leak Secrets,
Run rm -rf,
and Expose PII.

Tatu intercepts every Claude Code action before it lands. Leaked AWS keys get blocked. Destructive commands get denied. PII gets flagged. Every event becomes audit-ready evidence — running on your infrastructure, not ours.

View on GitHub See the Platform →
Apache 2.0 Licensed
0
Pre-built security policies
0
Frameworks mapped (SOC2, GDPR, LGPD...)
0
Agent lifecycle events captured
0
Your infrastructure, your data

Every secret leak, every destructive command, every PII exposure — blocked and logged

Claude Code hooks generate the telemetry. Tatu ingests, correlates, and blocks in real time. Your GRC team gets audit evidence. Your devs get guardrails. All on your own infrastructure.

Real-Time Threat Feed

An AWS key hits a config file — you see it in milliseconds, not in a post-mortem. Filter by severity, hook, developer, or repo.

Hook Telemetry

Which hooks fire most? Which ones never trigger? Tune your security policies with data, not guesswork.

Continuous Compliance Evidence

Every blocked action maps to SOC2, GDPR, LGPD, CPS234, and ISO 27001 controls automatically. No more spreadsheet scrambles before audits.

Team-Wide Patterns

See which agent behaviors trigger blocks most often across your org. Identify gaps in hook coverage before they become incidents.

Tamper-Evident Audit Trail

Structured JSON logs for every agent action. Your SIEM already speaks this language — forward via CSV, JSON, or direct integration.

Self-Hosted & Open Source

No vendor lock-in. No data leaving your network. Read every line of source code. Apache 2.0 — fork it, extend it, audit it yourself.

One dashboard for DevSecOps and GRC

Engineers see which hooks fire and why. CISOs see compliance posture across every repo. GRC analysts export audit evidence without asking anyone. Same data, different views — on your infrastructure.

localhost:3800/dashboard
EVENTS OBSERVED
7,209
THREATS BLOCKED
103
LIVE AGENTS
5
SECRETS INTERCEPTED
34
AGENT ACTIVITY TIMELINE
THREAT FEED
AWS key detected in config.py
rm -rf / blocked in prod branch
CPF pattern in test fixtures
tatu — audit trail
FULL OBSERVABILITY TRAIL
TIMESTAMP DEV HOOK EVENT RESULT
14:58:23 carlos.m Secrets Leak AWS key in config.py DENY
14:52:01 ana.r Cmd Blocker rm -rf / in infra-tf DENY
14:46:14 pedro.s PII Detector CPF in tests/ DENY
14:37:42 maria.l Dep Vuln CVE-2026-1847 WARN
14:29:18 julio.df Scope Check nmap validated PASS
Export CSV Export JSON Send to SIEM
tatu — compliance posture
SOC2 79%
GDPR 85%
LGPD 88%
CPS234 83%
ISO 27001 78%
CONTROL MAPPING
Session Audit Logger SOC2 CC7.2 · CPS234 · GDPR Art.30
Secrets Prevention SOC2 CC6.1 · LGPD Art.46 · GDPR Art.32
PII Detector LGPD Art.37/46 · GDPR Art.25 · SOC2 CC6.5

Three lines of config. Zero blind spots.

01

Clone and deploy

git clone, docker compose up. Under 5 minutes from zero to a running dashboard. Your data never touches an external server.

02

Point your hooks

Add one URL to .claude/settings.json. Every Bash command, file write, and tool call streams to your Tatu instance — PreToolUse, PostToolUse, all of them.

03

Observe, block, prove

Threats get blocked before they land. Compliance evidence accumulates with every session. When auditors ask, you export — not scramble.

// .claude/settings.json { "hooks": { "PreToolUse": [{ "matcher": "Bash|Write|Edit", "hooks": [{ "type": "http", "url": "http://localhost:3800/ingest" }] }] } }

Stop scrambling before audits

Every blocked secret, denied command, and flagged PII maps to compliance controls automatically. When your auditor asks 'how do you govern AI agent access?', you hand them a live dashboard — not a spreadsheet from last quarter.

SOC2

CC6.1 access controls, CC7.2 system monitoring, CC7.3 change detection, CC8.1 change management — covered by hook telemetry

GDPR

Art.25 data protection by design, Art.30 processing records, Art.32 security of processing — PII detection as evidence

LGPD

Art.37 processing records, Art.46 technical security measures — continuous DPO evidence from every agent session

CPS234

Information asset controls, mandatory logging, change management obligations — all generated from hook activity

ISO 27001

A.9.4 access control, A.12.4 event logging, A.12.6 vulnerability management, A.14.2 secure development — mapped automatically

Right now, your AI agents run unmonitored.
That ends with one command.

Free. Open source. Self-hosted. No signup, no credit card, no vendor call. Clone, deploy, and know exactly what every agent does — before your auditor has to ask.

$ git clone github.com/laboratoriohacker-com/tatu && docker compose up