Whether AiSOC is worth the hassle depends on who you are: if you handle a pile of security alerts every day, it is worth an afternoon to try; if you are just an individual user, it is not for you. AiSOC is an open-source AI security operations center that runs on a local model with no API keys at all. As of November 18, 2026, it had passed 8,100 GitHub stars.
Official repository
The official repository lives on GitHub, under the organization beenuar, project name AiSOC, under the MIT license, created on May 2, 2026. It works in four steps: first it takes in security telemetry, then it filters alerts with 2,966 executable rules drawn from a 7,408-rule library, then it groups related alerts into incidents, and finally an AI agent investigates each one, recording every prompt and tool call in a replayable investigation ledger. It connects to 97 data sources, including Splunk, Elastic, CrowdStrike, Okta, AWS, and Kubernetes audit logs. Automated actions require human approval by default, and three switches must all agree.
Deployment cost: a low bar, but the machine must be up to it
The deployment requirements are stated plainly: Docker Compose v2, 8 GB of memory, 20 GB of free disk, and a roughly 2 GB local model download on first run. The flow is to clone the repository, run the installer, then use the built-in smoke test to send one real event through the whole pipeline, with clear pass-or-fail feedback at every step. Once running, there is no usage-based bill, data never leaves your network, and it ships with an MCP server so Claude, Cursor, and Continue can query the security center directly. You can also skip real data at first and load sample data to watch five scenarios travel the same pipeline. Like any self-hosted open-source project, OpenClaw's deployment cost and security pitfalls are worth reading first, so you do not underestimate the operations work.
Three pitfalls: stars are not deployments
First, a local model saves the API bill, but it is smaller than frontier cloud models, so complex investigations lose capability. For comparison, Tencent's Hunyuan Hy4 preview is a 770B-parameter flagship model that needs a multi-GPU server to run; AiSOC takes a different path: a small model, running locally, good enough. Second, the project is only six months old, and its README admits that executability 'is earned, not declared': 2 of its 7,408 rules still cannot fire, so maturity needs verifying with the smoke test. Third, stars measure attention, not deployments; with no real data source connected, it is an empty shell. It suits small teams with security telemetry and developers who want to study AI-led investigation, not individual users with no data source. The alternative is to run the sample data first, then decide whether to connect real alerts.
Q: Can an individual without a security team use AiSOC?
A: It can be installed, but it is not recommended. AiSOC's value lies in handling real alerts; with no data source there is nothing to investigate, so try the sample data to see the workflow first.