Frequently asked questions

Jutsu is an AI-driven, autonomous Security Operations Center (SOC) platform. It ingests security events, enriches them with threat intelligence, uses AI to triage and correlate them into incidents, and automatically executes containment, escalating ambiguous cases to human analysts.

A traditional SIEM collects and searches logs. Jutsu sits on top of your SIEM and acts on the alerts — it triages each one with an LLM analyst, enriches it, correlates related alerts into incidents, runs automated response playbooks, and writes the reports. It is your Security Operations OS, not just another log store.

Four products that work together: AgentSOC (the core autonomous SOC), Red Team (continuous MITRE ATT&CK attack simulation and detection validation), Exposure (a free agentless external security assessment), and Firehose (high-volume SIEM load and throughput testing).

No. Jutsu automates the repetitive Level 1 and 2 work — triage, enrichment, correlation, and documentation — while analysts stay in control. Ambiguous or failed-containment alerts auto-escalate to human L2/L3 investigation cases, and every action is logged and auditable.

From Wazuh (via a webhook and a high-throughput Go forwarder that tails alerts.json at up to 20–50k events per second), syslog/beacon sources, and Google Workspace email logs. Splunk, Microsoft Sentinel, and CrowdStrike are on the roadmap as additional sources.

Yes. The platform includes a built-in alert generator that simulates phishing, brute force, C2 beacons, lateral movement, malware, exfiltration, privilege escalation, and ransomware in Wazuh, Splunk, or generic formats, so you can validate detection and response end to end.

A multi-call structured LLM acting as an L2/L3 analyst assigns each alert a summary, category, severity, a 0–100 risk score, a confidence score, and a verdict of true positive, false positive, or not sure — with full reasoning. Only true positives — and ambiguous alerts that clear a confidence threshold — move on to incident correlation.

OpenAI, Anthropic Claude, and Google Gemini are all supported. Each pipeline function — triage, enrichment, correlation, response, incident reporting, org reporting, and Copilot — has its own configurable model slot, with an automatic fallback if a model's quota is exhausted.

Through its built-in AgentSOAR engine or Shuffle, it can block an IP, isolate a host, power off an instance (AWS/Azure/GCP/Hostinger), disable a user, and block an email sender or domain (Google Workspace / Microsoft 365). If no playbook matches or no credential owns the target, it escalates to a human L2 case.

The Copilot is a chat assistant grounded in your own alerts and incidents. You select the relevant alert or incident IDs and it answers questions and uses tools against that context, streaming responses live in the dashboard.

A dedicated agent groups related alerts into incidents using deterministic key-matching and entity normalization, then applies multi-hop attack-chain playbooks (such as lateral movement and malware origin) with an LLM tie-breaker, before generating reports and orchestrating response.

No. Every verdict carries written reasoning, ambiguous and high-risk cases are escalated to human analysts, and all worker processing and API mutations are recorded in an audit log. AI token usage is also tracked per model and organization for full cost visibility.

The core AgentSOC is self-hosted and containerized — it runs via Docker Compose, Dokploy, or Kubernetes (with KEDA queue-depth autoscaling on GKE). Your data lives in your own PostgreSQL, OpenSearch, RabbitMQ, and S3. Red Team and Firehose are delivered cloud-first, and Exposure is an agentless hosted assessment.

Login is via email/password or Google SSO, with role-based access (user, analyst, admin, and super-admin) and admin actions gated to privileged roles. Tenancy is per-organization with row-level isolation, and a database-per-organization, multi-region design is available for deployments that require data residency.

The Go forwarder only tails your existing alerts.json file and reads its API key from the ossec.conf integration block. It uses at-least-once delivery with a single offset file, needs no spool, and starts at end-of-file so it never replays your alert history.

Up to nine during enrichment: VirusTotal, AbuseIPDB, GreyNoise, OTX (AlienVault), MISP, MalwareBazaar, Kaspersky OpenTIP, the CISA KEV catalog, and ip-api.com for geolocation. Results are cached in PostgreSQL to limit repeat lookups.

Yes. Each organization's data is isolated, webhook ingest is authenticated with per-source API keys, and service-to-service routes use a separate service key. A multi-region, database-per-organization topology is available for deployments that require data residency.

Exposure uses only public, non-invasive external signals and is explicitly not a penetration test — it runs no exploit payloads, credential tests, or scans of private IP ranges. Any secrets found in public content or breach data are masked before storage and never logged in clear.

No. Jutsu ingests logs, alerts, and telemetry from your current stack — there is no rip-and-replace. It connects to Wazuh, syslog, and Google Workspace today, with Splunk, Microsoft Sentinel, and CrowdStrike shown as coming-soon sources.

For containment it integrates with AWS, Azure, GCP, and Hostinger for instance control, and with Google Workspace and Microsoft 365 to disable users and block email senders or domains. Containment is routed to whichever credential owns the target resource.

Yes. LLM providers and threat-intel providers are configured per deployment with your own keys. Missing keys degrade gracefully — for example, if no LLM key is present, the response agent simply escalates to a human L2 case.

Red Team emits attack telemetry to AgentSOC via a Wazuh-style webhook ingest URL and API key, and supports Wazuh (Manager API and Indexer). Its generator can format alerts as wazuh, splunk, generic, json, or syslog; Splunk, CrowdStrike, Jira, and Slack are on the roadmap.

Beyond a TCP syslog listener and firewall access, you install the shipped local_decoder.xml and local_rules.xml on the Wazuh Manager so the synthetic events fire rules. Firehose then reads alerts back via the alerts.json file or the Wazuh Indexer to measure latency.

Connect your existing sources and the agents begin triaging immediately — there is no lengthy deployment project. You can also run the built-in alert generator to validate the pipeline before pointing real traffic at it.

Red Team launches realistic, MITRE ATT&CK-mapped attack simulations — brute force, malware, phishing, C2, ransomware, lateral movement, and full APT chains — against scoped lab targets, then correlates the results against your SIEM telemetry to validate whether your defenses actually detected and responded.

It is safe by design. Runs are scoped to lab/staging/prod environments with safety profiles, destructive scenarios are blocked outside permissive lab environments, prod or destructive launches require two-person approval, unknown targets are rejected, and a global kill switch can stop everything instantly.

Yes. Red Team correlates each run against your SIEM telemetry, recording which expected signals were detected versus missed and computing detection rate plus median and p95 mean-time-to-detect.

Exposure is a self-service, agentless external assessment. You enter your domain and email and it produces a board-ready report scoring nine risk dimensions — external exposure, attack surface, vulnerabilities, dark web, identity, cloud, program maturity, SOC readiness, and AI readiness — with OWASP mapping and a likely attack path.

Firehose is a synthetic event load generator and SIEM stress-tester for Wazuh. It pushes high-volume TCP syslog across realistic attack scenarios, reads the alerts back, and measures real ingest throughput (events per second) and detection latency (MTTD p50/p95/p99) per run.

No. Adaptive mode is a statistical simulator using Markov-style scenario transitions, a small consistent attacker pool, and a severity ramp. It is deliberately not an LLM and is not trained on your data.

Jutsu offers Free, Startup, Business, and Enterprise/MSSP plans. The first 5 assets are always free, the Startup plan is priced per asset, and Business and Enterprise are custom-quoted to your environment, asset volume, and compliance needs.

Yes. Start on the Free plan with no credit card, or run the free Exposure external assessment in minutes. You can also join the Design Partner Program to evaluate the platform in your own environment.

Notifications can be delivered via Email (SendGrid), Slack, Telegram, and PagerDuty. Incident reports are generated as Markdown and stored in S3, and an organization-reporting job produces weekly or daily SOC activity reports with MTTD/MTTR/MTTA metrics, rendered to HTML or PDF.

Early-stage startups can start free, growing startups use the Startup plan for 24/7 coverage without a SOC team, mid-market and regulated companies move to Business for compliance and multi-user access, and enterprises and MSSPs use the Enterprise plan for multi-tenant management and private deployment.

No. The platform is designed to give startups 24/7 autonomous coverage without standing up a SOC team. The AI agents handle triage, enrichment, and containment, and only escalate to a human when judgment is genuinely needed.

Yes. Webhook receipts, worker processing, and API mutations are written to a PostgreSQL audit log, logins are audited with IP, and Red Team additionally keeps an append-only, SHA-256 hash-chained audit log with a verify endpoint that detects tampering.