Nemorix - AI SRE Agent with Persistent Memory

• 

  Persistent Memory Graph

• 

  Incident Investigation Pipeline

• 

  Dashboard

Inspiration

Production incidents often repeat, but SREs still have to rediscover what caused them. I built Nemorix around one idea: an AI SRE agent becomes more useful when it remembers previous incidents. Nemorix gives AI persistent operational memory using CockroachDB, allowing historical incident knowledge to be retrieved and reused during new investigations.

What it does

Nemorix is an AI SRE agent with persistent operational memory. When an incident is selected, Nemorix analyzes its context, generates an embedding, searches historical memories in CockroachDB, and provides relevant incidents to the AI for contextual reasoning.

Workflow:

Incident → Embedding → CockroachDB Vector Search → Historical Memory → AI Reasoning → Runbook → Persistent Memory

In my demo, a new payment-svc incident retrieves similar historical incidents, including an 86% similarity match, before presenting investigation and remediation guidance.

How we built it

I built Nemorix using: React + Vite — frontend Node.js + Express — backend CockroachDB — persistent incident memory and vector search Amazon Bedrock — AI reasoning Amazon Nova Lite — reasoning model

CockroachDB connects current incidents with previous operational experience, making the AI workflow memory-grounded rather than stateless.

Challenges we ran into

The biggest challenge was making historical incidents useful to the AI rather than simply displaying old records. I built a semantic retrieval workflow connecting structured incident data, embeddings, and CockroachDB vector search. I also designed the interface to make the investigation pipeline visible to an SRE.

Accomplishments that we're proud of

I am proud that Nemorix demonstrates a complete operational memory loop:

Store → Retrieve → Reason → Act → Remember

Nemorix can persist incidents, retrieve similar failures, show similarity scores, provide historical context to the AI, and generate remediation guidance.

What we learned

I learned that effective agentic memory is not about storing everything. It is about retrieving the right knowledge at the right time. I also learned that CockroachDB can play an active role in an AI agent architecture by connecting current incidents with historical operational experience.

What's next for Nemorix — AI SRE Agent with Persistent Memory

My next goal is to make Nemorix a more complete production SRE companion. I plan to connect additional signals such as deployments, code changes, and metrics; improve memory and runbook quality; add stronger observability; and eventually support controlled automated remediation with human approval.

Built With

• agentic-ai
• ai
• amazon-bedrock
• amazon-nova-lite
• aws-sdk
• cockroachdb
• cockroachdb-cloud-mcp
• cockroachdb-vector-indexing
• express.js
• javascript
• node.js
• persistent-memory
• react
• semantic-search
• sre
• vector-search
• vite