Self-hosted AIOps with local models
Logs hold customer data, secrets and internal hostnames. Sending them to a hosted AI is a non-starter for many teams. SerOps runs on your own infrastructure and can use a local model through Ollama, so incident analysis never leaves your network.
Everything runs where you put it
SerOps is open source (AGPL-3.0) and self-hosted: Docker Compose on a single host, or Kubernetes with the official Helm chart. The application, its PostgreSQL and Redis, and the analyses all live on your infrastructure. It queries your log and metric backends inside your network and keeps no long-term copy of your telemetry.
A local model with Ollama
SerOps has a first-class Ollama integration: the same pipeline, the same prompt and the same structured output as the hosted models, with a model running on your own hardware. Point SerOps at your Ollama endpoint, pick the model, and every scheduled analysis, investigation and chat answer is produced locally.
You can still compare models on the same incident, for example a local model against a hosted one on non-sensitive environments, and keep the one that gives the best root causes for your workload.
Masking before analysis, whatever the model
Before any model sees a log line, SerOps masks secrets and personal data: passwords, API keys and tokens, private keys, connection strings, emails, phone numbers and card numbers. That applies to hosted and local models alike. See data protection and PII redaction.
Built for regulated teams
- Single sign-on with OIDC providers such as Keycloak, Okta, Azure AD or Google, and admin, user and viewer roles. See authentication.
- Audit log of logins and changes, with retention you control. See compliance.
- Multi-tenant organisations with strict data isolation. See multi-tenancy.
- No telemetry phoned home. SerOps sends nothing to us.
Frequently asked questions
Can SerOps run fully air-gapped?
Yes, with a local model through Ollama and container images mirrored to your own registry; both the Compose file and the Helm chart let you set the image location.
Which local models work?
Any model you run in Ollama. Larger models generally give better root causes; compare them on your own incidents with model comparison.
Can I mix local and hosted models?
Yes. Choose a primary model and re-analyse specific incidents with another one when you want a second opinion.
What does it cost?
Nothing for SerOps itself: every feature is free and open source when you self-host. With a local model there is no per-call AI cost either.
See it on real data
The live demo runs on seeded incidents: no signup. Or run SerOps on your own infrastructure; it is free and open source (AGPL-3.0).