AiSOC

Open source alternative to Palo Alto Networks Cortex XSIAM

6.2K stars
MIT
Active this week
Last commit today

Last verified Nov 2, 2026

Monitor, detect, and respond to security threats with a self-hostable AI SOC. Every agent decision is logged to an auditable Investigation Ledger; MIT licensed.

Visit website GitHub6.2K Deploy on Hostinger

Read the full breakdownWho it is for, how it works, trade-offs, the quick start and the FAQ

who it's for

Who AiSOC is for#

Enterprise security teams with SOC 2 or DORA compliance requirements

Teams in SOC 2, ISO 27001, PCI-DSS, or DORA-regulated environments need to document every automated closure decision. AiSOC's Investigation Ledger records every agent step with its evidence citations, exportable as a signed bundle for auditors. Built-in compliance dashboards track MTTD, MTTR, and MTTC SLA metrics against each framework's controls.

Skip if:

Your compliance requirement does not extend to AI triage decisions. If human analysts review every alert before closure, a traditional SIEM without an AI layer may cover compliance needs at lower operational cost.

Blue teams running detection engineering programs

Security engineers who write and maintain detection rules can use the Sigma-over-OpenSearch pipeline with the 7,023-rule community catalog and one-click install. The purple-team module pairs Atomic Red Team and Caldera execution with MITRE ATT&CK heatmap coverage so teams can validate detection coverage against their actual rule library.

Skip if:

Your team only needs alert visualization and does not maintain custom detection rules. A lighter SIEM with pre-built dashboards is faster to stand up if detection engineering is not a core function.

Security teams in air-gapped or data-restricted environments

The core stack runs on a single server with no external API calls beyond the CISA KEV feed (public, no key). The bundled Ollama model produces triage verdicts locally. Teams that cannot send security telemetry to cloud vendors for regulatory or data-classification reasons can run the full AI triage pipeline on their own hardware with no outbound connection required.

Skip if:

Your environment allows cloud AI providers and you need frontier-model triage quality. Configuring AiSOC against a cloud OpenAI-compatible endpoint is the better option when data residency is not a constraint.

Security engineers evaluating AI SOC tooling for regulated enterprises

The public 200-case eval suite is reproducible on a laptop in under 30 seconds and documents its methodology including what each metric measures and what it does not. Every verdict scorecard names the model that actually served it. This makes AiSOC one of the few AI SOC tools where a buyer can independently verify triage quality claims before committing.

Skip if:

You need a vendor-supported, fully managed platform with SLA guarantees and professional services. AiSOC is maintained by the open source community; production support comes from the team running it internally.

the problem

The problem it solves#

Security operations teams accumulate thousands of alerts per day from EDR, SIEM, cloud, and identity tools. Triage takes most of the analyst's day, but the harder problem is documentation: when an auditor asks why an alert was closed, an analyst pointing to an AI verdict with no reasoning trail has nothing to show.

For regulated organizations, this opacity is a blocker. SOC 2, ISO 27001, and DORA controls require evidence that automated decisions are documented and reviewable. Cloud AI SOC platforms keep their agent logic proprietary, which means every AI-assisted closure is a potential audit gap. Keeping humans in the loop for every decision does not scale when alert volumes double each quarter.

how AiSOC solves it

How it solves it#

Investigation Ledger with step-by-step replay

Every prompt, tool call, citation, and token cost the AI agent produces is persisted to the Investigation Ledger and replayable step by step in the case workspace. Investigations export as signed OCSF 1.9.0 evidence bundles with prompt digests, so an auditor can verify a closure against the original evidence without trusting a summary.

7,023-rule detection engine with 2,538 verified executable rules

The engine runs Sigma over OpenSearch and ClickHouse, YARA, and KQL/EQL against a library of 7,023 rules. A rule is counted as executable only when a vendor-shaped event has been replayed through the real connector and the rule was observed to fire; inference from a directory listing does not count. Two rules remain unexecutable, down from 119 in earlier versions.

LangGraph multi-agent investigation pipeline

Recon, forensic, responder, and reporter agents wire through a LangGraph orchestrator to triage alerts and enrich cases. The model picks typed tools for lake queries, graph traversals, and enrichment lookups; it never writes SQL directly. A grounding check demotes a verdict citing an indicator not found in the evidence to human review rather than auto-closing.

90 data connectors with detection migration tooling

Settings lists 90 click-and-connect data connectors including CrowdStrike, Splunk, Sentinel, Elastic, Okta, AWS, and Kubernetes audit with vendor-specific normalization. The `aisoc-migrate` package translates Splunk SPL, Sentinel KQL, and Elastic EQL rules and explicitly refuses to approximate what it cannot carry, so untranslatable rules surface as missing rather than silently wrong.

Compliance dashboards for six frameworks

Built-in dashboards track SOC 2, ISO 27001, NIST CSF, PCI-DSS, HIPAA, and DORA evidence with MTTD, MTTR, and MTTC SLA metrics. The compliance reports draw from the same investigation data that feeds the audit log, not a separate reporting path.

Playbook engine with visual editor and three-switch authorization

A React Flow visual editor ships 12 starter playbook templates for automated and human-gated response actions. Three switches must agree before a playbook executes against a real target: the deployment-level switch, the tenant-level switch, and the playbook-level switch. All three default to off. Approval steps are durable pauses stored in Postgres that survive restarts.

strengths · trade-offs

Strengths and trade-offs#

Strengths

Fully auditable agent reasoning trailEvery LLM prompt, tool call, citation, and token cost is written to the Investigation Ledger. Any investigation exports as a signed evidence bundle, byte-identical, with prompt digests and OCSF 1.9.0 encoding. This is the specific gap that commercial AI SOC platforms leave open: Cortex XSIAM runs a closed agent loop on vendor infrastructure with no published reasoning trail available for audit.
Runs fully offline with a bundled local modelThe core stack ships Ollama with `llama3.2:3b-instruct-q4_K_M` pinned in the gateway. On 50 alerts that model gave usable triage output on 50 after constraining the reply to JSON, as documented in the repository's measurement script. No external API endpoint is called. Security teams with air-gapped networks or cloud-data restrictions can run the full AI triage pipeline on their own hardware.
MIT license with no telemetry or calls homeThe code, prompts, and templates are MIT licensed with no CLA and no phone-home behavior. Organizations can inspect the agent's full prompt library, modify detection templates, and run the stack commercially without a vendor agreement. There is no hosted-service restriction because MIT carries none.
Public 200-case CI regression suite on every pull requestA 200-incident CI-gated regression suite runs on every pull request targeting main or develop. The suite methodology is published and reproducible on a laptop in under 30 seconds. Teams can verify that detection quality has not regressed before merging, and the methodology explains what each metric measures and what it does not.

Trade-offs

-High resource requirements for the full stackThe core profile needs 8 GB RAM and 20 GB free disk in the Docker VM, plus Docker Compose v2 and Python 3.9+. The full profile, which adds the event lake, entity graph, and full-text search, needs 12 GB RAM and runs 22 services. Teams on a standard 2 GB or 4 GB VPS will not have enough headroom; a dedicated server or a cloud instance sized at 8 GB+ is the practical minimum.
-Bundled model is not a frontier LLMThe shipped `llama3.2:3b-instruct-q4_K_M` model is sized for CPU-only inference and is not comparable to GPT-4o or Claude for nuanced investigation reasoning. The repository documents triage reliability at 50/50 on constrained JSON output over a 50-alert sample. Teams requiring higher-quality triage for complex incidents will need to configure an external OpenAI-compatible endpoint, which requires either a cloud provider account or a local GPU.
-Not all 7,023 detection rules are executableOf 7,023 rules in the library, 2,538 are verified executable against real vendor-shaped events. 1,362 Sigma rules were refused during compilation, documented in `docs/detections/sigma-compilation.md`. Teams migrating from Splunk will find 271 of 2,005 bundled SPL rules do not translate. The repository documents what does not translate and why, but the gap is real and should be reviewed before relying on a specific rule set.

versus alternatives

AiSOC vs alternatives#

AiSOC vs Palo Alto Networks Cortex XSIAM

Both tools use AI agents to triage security alerts, but they target different deployment models and governance requirements. XSIAM is a cloud-delivered, vendor-managed platform; AiSOC runs on your own infrastructure with the full agent loop visible and auditable.

Feature	AiSOC	Cortex XSIAM
License	MIT	Proprietary
Deployment	Self-hosted (Docker Compose, Helm)	Cloud-only
Agent reasoning	Auditable via Investigation Ledger	Closed agent loop
Local model	Yes, bundled Ollama (no API key needed)	Vendor-managed
Compliance dashboards	SOC 2, ISO 27001, NIST CSF, PCI-DSS, HIPAA, DORA	Vendor-reported
Detection engine	7,023-rule library, 2,538 executable	Proprietary catalog
Pricing	Free to self-host (MIT)	Per-endpoint commercial licensing

AiSOC is the stronger choice when an organization needs to document the AI triage decision trail for compliance audits, keep security telemetry on its own infrastructure, or inspect and modify the detection engine directly. For regulated teams where every automated closure must cite its evidence, AiSOC's signed investigation bundles and replay capability address a gap that XSIAM's closed model cannot close.

XSIAM remains worth considering for large enterprise SOCs that want a fully managed platform with Palo Alto's threat intelligence network, vendor-supported integrations at scale, and no infrastructure management overhead. Teams without dedicated security engineers to operate and size the Docker Compose stack will find the managed path materially simpler, even at commercial pricing.

install · quick start

Quick start#

bash

Self-hosting AiSOC uses Docker Compose v2; the install script handles prerequisites and starts the full stack on Linux and macOS.
```bash
curl -fsSL https://raw.githubusercontent.com/beenuar/AiSOC/main/install.sh | bash
```

tech stack · detected from GitHub

What it's built on#

Languages: GoPythonTypeScript
Frameworks: LangGraphNext.jsReact

frequently asked

FAQ#

Does AiSOC require an OpenAI or cloud API key to work?

No. make up ships a bundled Ollama model (llama3.2:3b-instruct-q4_K_M) that runs entirely on CPU. The core stack produces real triage verdicts with no external API key on first boot. You can configure an external OpenAI-compatible endpoint from the console if you want higher-quality model output; the bundled model is the default for data-restricted or air-gapped environments.

How many detection rules does AiSOC actually execute?

The library contains 7,023 rules. Of those, 2,538 are verified executable: a rule reaches that count only when a vendor-shaped event has been replayed through the real connector and the rule was watched to fire. Two rules are not yet executable (down from 119 in earlier versions) because they are waiting on a behavioral baseline. 1,362 Sigma rules were refused during compilation; the full list and reasons are in docs/detections/sigma-compilation.md.

What are the minimum server requirements to run AiSOC?

The core profile (ingest, detect, correlate, alert, AI triage, console) needs 8 GB RAM, 20 GB free disk in the Docker VM, Docker Compose v2, Python 3.9+, and bash. make doctor checks all requirements before install. The full profile adds the event lake, entity graph, and full-text search and needs 12 GB RAM. GitHub Codespaces works as an alternative if you do not have a suitable local machine.

How does AiSOC prevent playbooks from taking unintended automated actions?

Three switches must all be on before a playbook executes against a real target: the deployment-level switch, the tenant-level switch, and the playbook-level switch. Every default is off. Anything less runs in preview with the proposed plan attached to the alert. Approval steps are durable Postgres-persisted pauses that survive restarts, and they expire with a recorded outcome rather than hanging indefinitely.

Can AiSOC import existing detection rules from Splunk, Sentinel, or Elastic?

Yes, via packages/aisoc-migrate. It translates Splunk SPL, Sentinel KQL, and Elastic EQL rules and explicitly refuses rather than approximating what it cannot carry. Of 2,005 bundled Splunk rules, 1,734 translate and 1,711 of those are partial (the documentation explains what partial means). The refusal-rather-than-approximate policy means untranslatable rules surface as missing rather than becoming silent gaps.

also worth a look

Similar open-source tools#

REA (Reverse Engineer Anything)

Featured

Alternative toRevEng.AI

AI-powered reverse engineering for binaries, apps, and assemblies.

418,752 GitHub stars
Licence MIT

Oya Browser

Alternative toBrowserbase

Records portal tasks once; replays with no model and no bot detection

351 GitHub stars
Licence Sustainable Use License 1.0

Agent Security

Alternative toLakera Guard

On-device policy enforcement and secret detection for AI coding agents

158 GitHub stars
Licence Apache-2.0

Promptfoo

Alternative toBraintrust

Automated LLM evals, red teaming, and AI vulnerability scanning.

26,680 GitHub stars
Licence MIT

iFixAi

Alternative toZenity

Self-hosted AI agent auditing across 60 trust and safety inspections.

37,967 GitHub stars
Licence Apache-2.0

Bifrost

Alternative toPortkey

Open-source AI gateway with 23+ providers, failover, and MCP support.

9,212 GitHub stars
Licence Apache-2.0

by category

Open source tools in these categories#

152

Security & Monitoring

REA (Reverse Engineer Anything), RustDesk, Uptime Kuma 450

AI & Machine Learning

freeCodeCamp, REA (Reverse Engineer Anything), OpenClaw

Hostinger

Self-host AiSOC on a Hostinger VPS

Keep it running 24/7, even when your laptop is off. A Hostinger VPS gives it a server of its own.

01
Choose a plan
Starts from $6.49/mo
02
Apply
Extra 10% off yearly plans
03
Install AiSOC
On a server that stays on
04
Start building
Focus on the app, not the server

Deploy on Hostinger

Inbox Zero

Alternative toSuperhuman

AI email assistant that drafts replies and clears your inbox

12,855 GitHub stars
Licence AGPL-3.0
#Email

rakazo

Alternative toGrok Bot

AI teammates you own: your keys, your model, your machine.

6,568 GitHub stars
Licence Apache-2.0
#AI Agents

SnitchFeed

Live monitoring on LinkedIn, X, Reddit & more. Agentic social listening for startups.

Marketing & Growth

Advertise your product

The projects we add and every new guide. Free, unsubscribe any time.

Find open source alternatives to paid software

AiSOC vs Palo Alto Networks Cortex XSIAM

Feature	AiSOC	Cortex XSIAM
License	MIT	Proprietary
Deployment	Self-hosted (Docker Compose, Helm)	Cloud-only
Agent reasoning	Auditable via Investigation Ledger	Closed agent loop
Local model	Yes, bundled Ollama (no API key needed)	Vendor-managed
Compliance dashboards	SOC 2, ISO 27001, NIST CSF, PCI-DSS, HIPAA, DORA	Vendor-reported
Detection engine	7,023-rule library, 2,538 executable	Proprietary catalog
Pricing	Free to self-host (MIT)	Per-endpoint commercial licensing