Komodor adds AI memory to its site reliability platform

Komodor has added Klaudia Memory to its AI site reliability engineering platform. The feature draws on customer-specific operational history during incident investigations.

The addition extends Klaudia, Komodor's AI agent for root cause analysis and remediation, with a system that retains context from earlier investigations. It can use patterns from previous incidents, customer-approved runbooks and architecture notes to guide later responses.

The company is targeting a long-running problem for site reliability engineering teams managing complex cloud-native systems. Engineers often rely on a mix of postmortems, undocumented fixes, compliance requirements and internal architecture decisions when responding to production failures, and much of that knowledge sits with a small group of senior staff.

Klaudia Memory is intended to capture that record automatically from prior investigations. It can retain correlations between symptoms and likely causes, record which remediation steps worked in earlier cases, and recognise repeated timing patterns tied to scheduled jobs or recurring workloads.

This approach is designed to help teams avoid retracing unsuccessful steps when a similar incident reappears. The software also becomes more selective over time by learning which alerts and signals are routine in a specific environment and do not need attention.

"Since every organization has its own architecture, dependencies, constraints, and history of how past incidents were resolved, generic AI co-pilots lack the context to automate the management of enterprise production environments," said Itiel Shwartz, co-founder and CTO of Komodor.

"Klaudia Memory gives teams a way to preserve that institutional knowledge and make it available during every investigation, without training on customer data or exposing one customer's information to another," Shwartz said.

Three layers

The new feature sits alongside two existing sources of context in the platform. One is a knowledge base integration that lets Klaudia refer to customer-supplied runbooks, postmortems and operational documents. The other is Klaudia.md, a layer intended to store system rules and constraints that are not obvious from infrastructure manifests alone.

According to Komodor, the knowledge base integration allows the AI tool to use approved internal procedures rather than generic troubleshooting paths. During an investigation, it can identify relevant sections in uploaded material and follow the steps described there while using the customer's own terminology.

Klaudia.md is designed to address another common issue in operations work: important production constraints are often known to teams but not recorded in machine-readable infrastructure definitions. These can include failover rules, scaling limits linked to licensing or cost controls, and compliance restrictions on how services can be changed.

Komodor gave the example of a machine learning service that should not be reduced below 16Gi of memory because it would fail on start-up even without traffic. By loading this context at the beginning of a session, the platform is intended to stop investigators from choosing a technically plausible fix that would create a new problem in production.

Wider access

The release also broadens how engineers can access Klaudia. Teams can now trigger investigations and interact with the system outside Komodor's own user interface, including from Slack and Microsoft Teams war rooms, developer tools such as VS Code, Claude Code and Cursor, and version control or GitOps workflows.

Application programming interface and model context protocol integrations allow customers to connect the tool to internal systems and bespoke workflows. The platform also now supports a wider set of automated follow-up actions, including one-click remediation, self-healing policies, pull requests, tickets and postmortems.

Komodor is also expanding the set of specialist agents used by Klaudia across different parts of the cloud-native stack. Those agents are meant to support investigation and remediation work across code, infrastructure and applications, including workloads that run outside Kubernetes on services such as EC2 and ECS.

The update comes as software suppliers race to embed AI systems more deeply into operations teams' workflows. A central question for buyers has been whether these tools can act reliably in production settings where undocumented dependencies, internal rules and service history matter as much as logs and telemetry.

Komodor, which says it has raised USD $90 million in venture funding, is positioning the latest release around that operational context. The new version of its platform is available immediately through the company and its channel partners worldwide.