AiSOC v20 AI Security Operations Center

AiSOC v20.3.0 is an open-source, self-hostable AI Security Operations Center covering telemetry ingest, detection, correlation, an event lake, an entity graph and threat intel. This template deploys its full production stack, about twenty services, with a bundled local model and human approval for response actions.

What you get

  • AiSOC v20.3.0 services: web console, API, ingest, fusion, agents, actions, connectors, threat intel, realtime, enrichment and UEBA
  • Data plane of Postgres, Redis, Redpanda (Kafka API), ClickHouse, OpenSearch, Neo4j and Qdrant, each with its own volume
  • LiteLLM gateway and Ollama with a local llama3.2 3B model (about 2 GB), so no API key is required
  • Generated secrets and a first administrator; only the console and the token-protected ingest endpoint are public
  • A setup wizard with sample data that runs five events through the real pipeline

Good for

  • A self-hosted SOC console that correlates alerts from EDR, SIEM, identity and cloud tools into incidents
  • AI triage of alert queues with recorded, auditable reasoning
  • Hunting over an event lake and entity graph enriched with public threat intelligence
  • Keeping response actions such as host isolation behind human approval

Services in this template

  • connectors · image ghcr.io/beenuar/aisoc-connectors:v20.3.0
  • qdrant · image qdrant/qdrant:v1.19.2
  • opensearch · from baranberkay96/aisoc-railway
  • ollama · from baranberkay96/aisoc-railway
  • neo4j · image neo4j:5.15.0-community
  • threatintel · image ghcr.io/beenuar/aisoc-threatintel:v20.3.0
  • web · image ghcr.io/beenuar/aisoc-web:v20.3.0
  • fusion · image ghcr.io/beenuar/aisoc-fusion:v20.3.0
  • enrichment · image ghcr.io/beenuar/aisoc-enrichment:v20.3.0
  • actions · image ghcr.io/beenuar/aisoc-actions:v20.3.0
  • clickhouse · from baranberkay96/aisoc-railway
  • agents · image ghcr.io/beenuar/aisoc-agents:v20.3.0
  • litellm · from baranberkay96/aisoc-railway
  • Redis · image redis:8.2.11
  • Postgres · image ghcr.io/railwayapp-templates/postgres-ssl:18
  • api · image ghcr.io/beenuar/aisoc-core-api:v20.3.0
  • ueba · image ghcr.io/beenuar/aisoc-ueba:v20.3.0
  • kafka · image redpandadata/redpanda:v24.3.18
  • realtime · image ghcr.io/beenuar/aisoc-realtime:v20.3.0
  • ingest · image ghcr.io/beenuar/aisoc-ingest:v20.3.0

Questions people ask

How do I log in?

Wait until api and web are healthy; the first deploy pulls about 8 GB of images. Open the web URL and sign in with AISOC_ADMIN_EMAIL (default admin@aisoc.internal) and AISOC_ADMIN_PASSWORD from the api service's Variables. The setup wizard can load sample data and mint a push token for the ingest endpoint.

Do I need an LLM key?

No. Triage runs on the bundled local CPU model. To use your own provider, set it per tenant in the console, or set OPENAI_API_KEY or ANTHROPIC_API_KEY and the model variables on the litellm service.

What drives the cost?

The number of services: about twenty, including ClickHouse, OpenSearch, Neo4j and a local LLM, each with memory and a volume. The template's own estimates, not measurements, are about $57 per month idle and about $150 with typical use. The Hobby plan is the stated minimum.

How do I scale or upgrade?

web, api, ingest, actions and enrichment can take more replicas; fusion and agents stay at one replica because they use volumes. To upgrade, change the tag on all eleven aisoc images together and redeploy; the pre-deploy step applies migrations.

Are there licence restrictions?

AiSOC is MIT-licensed. Third-party parts have their own terms: Redpanda Community Edition is under the Business Source License 1.1 (free for self-hosted use, not for offering Redpanda as a managed service), and the llama3.2 model is under the Llama 3.2 Community License.

Full template README

Deploy and Host AiSOC with Railway

AiSOC is an open-source, self-hostable AI Security Operations Center. This template deploys its full v20.3.0 production stack on Railway in one step. It covers telemetry ingest, detection, correlation, an event lake, an entity graph and threat intel. AI triage runs on a bundled local model, and actions that change vendor systems wait for human approval. No API key is required.

About Hosting AiSOC

AiSOC takes security telemetry from your tools through an authenticated ingest API or 80+ connectors and normalizes it. Fusion matches it against a large detection library and groups what fires into incidents. AI agents then triage those incidents and record every prompt and tool call. Proposed responses go to an approval queue.

Hosting it means running about twenty cooperating services: the console, API and workers, a Kafka-API broker, Postgres, Redis, ClickHouse, OpenSearch, Neo4j, Qdrant, and an LLM gateway with a local model. This template wires them on Railway's private network. It generates every secret and creates the first administrator. Only the console and the ingest endpoint are public. This is a community template and is not affiliated with or endorsed by the AiSOC project.

Common Use Cases

  • Run a self-hosted SOC console that ingests alerts from EDR, SIEM, identity and cloud tools, then deduplicates and correlates them into incidents.
  • Triage alert queues with AI agents whose reasoning is recorded and auditable, using a local model or your own LLM provider.
  • Hunt over an event lake (ClickHouse) and an entity graph (Neo4j), enriched with public threat intelligence such as the CISA KEV catalogue.
  • Keep response actions such as host isolation and account disablement behind an explicit human approval step.
  • Evaluate AiSOC end to end with the built-in sample data before you connect production sources.

Dependencies for AiSOC Hosting

  • AiSOC v20.3.0 images (web, api, ingest, fusion, agents, actions, connectors, threatintel, realtime, enrichment, ueba)
  • PostgreSQL 18 (Railway), Redis 8.2
  • Redpanda (Kafka API) as the event spine
  • ClickHouse 23.8 (event lake), OpenSearch 2.11 (IOC index), Neo4j 5.15 (entity graph), Qdrant 1.19 (vectors)
  • LiteLLM gateway and Ollama with llama3.2:3b-instruct-q4_K_M (local CPU model, about 2 GB)

Deployment Dependencies

Implementation Details

Service Role Public
web Next.js console; proxies the API, agents and realtime on the private network yes
ingest Telemetry endpoint (/v1/ingest, /v1/inbox/*), token required yes
api Core API; pre-deploy runs the SQL migrations and creates the administrator no
fusion Detection engine, correlation, alerts, ClickHouse lake writer (volume for its spool) no
agents AI triage, investigations, copilot, hunts, playbooks (volume for playbooks) no
actions Governed response actions with human approval no
connectors Vendor pollers and federated search; credentials encrypted at rest no
threatintel CISA KEV, OTX and MISP feeds into Qdrant and OpenSearch no
realtime Live updates over WebSocket/SSE no
enrichment, ueba IOC enrichment; behavioural baselines no
litellm, ollama LLM gateway and local CPU model no
kafka, Postgres, Redis, clickhouse, opensearch, neo4j, qdrant Data plane, each with its own volume no

First login

  1. Wait until api and web are healthy. The first deploy pulls about 8 GB of images and builds four small wrapper images. The local model downloads in the background.
  2. Open the web URL and sign in with AISOC_ADMIN_EMAIL (default admin@aisoc.internal) and AISOC_ADMIN_PASSWORD, found under the api service's Variables.
  3. In the setup wizard, press Load sample data. Five labelled events go through the real pipeline and become alerts with AI triage.
  4. Mint a push token in the wizard and send telemetry to https://{ingest-domain}/v1/ingest with Authorization: Bearer {token}, or add connectors in the console.

Using your own LLM: set it per tenant in the console (Settings → Deployment & AI). For the whole deployment, set OPENAI_API_KEY (or ANTHROPIC_API_KEY) and the two AISOC_LLM_MODEL_* variables on litellm, then clear AISOC_LLM_API_BASE.

Scaling: web, api, ingest, actions and enrichment can take more replicas. fusion and agents keep one replica because they use volumes, so give them more memory instead. Kafka topics are created with one partition; add partitions before you scale consumers.

Versions and upgrades: all AiSOC images are pinned to v20.3.0. To upgrade, change the tag on all eleven ghcr.io/beenuar/aisoc-* images together and redeploy. The pre-deploy step applies new migrations and fails the deploy if one fails. Third-party images are pinned by tag or digest.

Cost: about $57/month idle and about $150/month with typical use (estimates). Hobby is the minimum plan. Pro is recommended for real telemetry volumes.

Why Deploy AiSOC on Railway?

Railway runs the full multi-service AiSOC stack without you having to manage servers or Docker hosts. The data stores sit on an encrypted private network, each with a persistent volume, and only the console and ingest endpoint are exposed over HTTPS. Secrets are generated at deploy time. You can scale web and API replicas, add memory to workers, or switch to a hosted LLM by changing variables.

View AiSOC v20 AI Security Operations Center on railway.com

Similar templates

TemplateCategoryDeploymentsDeploys
9Router
Self-hosted AI router for Claude, Gemini, Cursor & OpenAI tools.
AI/ML9,734
flowise-railway
Flowise - low-code LLM apps builder
AI/ML8,512
LibreChat
Open-source ChatGPT clone: multi-model, agents, search, multi-user auth
AI/ML6,762
Hermes Agent by Nous Research
Template for Hermes Agent by Nous Research
AI/ML6,413
OpenClaw
[Homebrew support] The AI that actually does things
AI/ML5,167
Hermes Agent
Self Host Hermes Agent. Autonomous AI agent w/ self-improving, tools & more
AI/ML4,092