swapnildahiphale/OpenSRE

OpenSRE — the memory-first AI SRE
Self-hosted · Episodic memory · Knowledge graph · Web console · Slack & Teams
_{opensre.in — not affiliated with Tracer / opensre.com}

OpenSRE is an open-source AI SRE agent that automatically investigates production incidents, finds root causes, and learns from every investigation. It combines episodic memory (remembering past incidents and what fixed them) with a Neo4j knowledge graph (understanding service dependencies and blast radius) and 51 production-ready skills for tools like Datadog, Grafana, PagerDuty, Elasticsearch, Kubernetes, and AWS. Self-hosted, direct Anthropic by default (provider-agnostic via the optional LiteLLM proxy), and licensed Apache 2.0.

_{Click to watch OpenSRE investigate an incident in 60 seconds}

Website · Docs · Live Demo · Contributing

Why OpenSRE?

Learns from every incident	OpenSRE remembers past investigations — what worked, what didn't. Similar incident at 3am? It already knows the playbook.
Understands your infrastructure	Neo4j knowledge graph maps service dependencies, so the agent knows blast radius before it starts investigating.
Plugs into what you already use	51 production skills for Datadog, Grafana, PagerDuty, Elasticsearch, Kubernetes, AWS, and more. No rip-and-replace.

Quick Start

⚠️ Local-dev mode — not for production exposure. make dev starts the agent in simple-mode (cleartext HTTP, no rate limiting, no sandbox isolation). It is designed for use on localhost only. Before exposing OpenSRE on any network, read the Simple-Mode / Local-Dev Security Posture and Hardening Checklist in SECURITY.md.

git clone https://github.com/swapnildahiphale/OpenSRE.git
cd OpenSRE
cp .env.example .env
# Add your ANTHROPIC_API_KEY to .env (OpenRouter via the optional LiteLLM proxy)
make dev

Requires Docker Desktop (or equivalent) running. make dev creates the external data volumes if they are missing, then starts Postgres, config-service, Neo4j, sre-agent, and the web console. Migrations run automatically. Open http://localhost:3002 and paste the admin token shown in the terminal to sign in. (LiteLLM is an optional --profile litellm add-on.)

Full setup guide · Entra SSO · Slack integration · Configuration

Architecture

→ Detailed architecture docs · Architecture overview

Features

Feature	Description
51 Production Skills	Elasticsearch, Datadog, Grafana, PagerDuty, K8s, AWS, Jenkins, Argo CD, and more
Long-term Memory	Neo4j episodic memory — surfaces past solutions mid-investigation via memory-search
Knowledge Graph	Neo4j service topology, dependency traversal, blast radius
Multi-provider LLM	Direct Anthropic by default; optional LiteLLM for OpenAI, Gemini, and more
Web Console	Investigations, memory hub, config editor
Entra SSO	Sign in to the web console with Microsoft Entra ID (token login remains)
Slack Integration	Investigate incidents directly from Slack (make dev-slack)
Teams Bot	Investigate incidents from Microsoft Teams (make dev-teams)

→ See all features · Roadmap

Useful Commands

Command	What it does
make dev	Start all services (Postgres, config, Neo4j, agent, web UI)
make dev-slack	Start all services + Slack bot
make dev-teams	Start all services + Microsoft Teams bot
make stop	Stop all services
make status	Show service health status
make logs	Follow all service logs
make logs-agent	Follow sre-agent logs only
make clean	Remove containers, volumes, and images

Slack integration

Create a Slack app, add SLACK_BOT_TOKEN and SLACK_APP_TOKEN to .env, and run make dev-slack. Full guide.

Microsoft Teams

Configure TEAMS_APP_ID, TEAMS_APP_PASSWORD, and TEAMS_TENANT_ID in .env, then run make dev-teams. The bot listens on port 3978. See teams-bot/README.md.

Web console Entra SSO

Create a confidential Web Entra app (redirect http://localhost:3002/api/auth/callback locally), paste tenant / client id into Admin → SSO, and set SSO_CLIENT_SECRET on config-service. Keep Helm services.webUi.oidc.enabled off — that PKCE path is separate. Token login remains as break-glass. Full guide · docs/SSO_SETUP.md.

E2E Testing with EKS

Run OpenSRE against a real Kubernetes cluster with the OpenTelemetry Demo app to test end-to-end investigations.

Prerequisites

• An existing EKS cluster with kubectl and helm installed
• AWS CLI configured with access to the cluster

Setup

export EKS_CLUSTER=my-cluster
export EKS_REGION=us-west-2
make e2e-setup-eks

This installs the otel-demo app on your EKS cluster, sets up port-forward tunnels to Prometheus/Grafana/Jaeger, starts sre-agent and the web UI, and generates a team token you can use to sign in.

Run fault injection tests

make e2e-test                    # Quick cart failure investigation (raw curl)
make e2e-test-cart               # Cart service fault — ~10% EmptyCart failures
make e2e-test-product            # Product catalog fault — ~5% GetProduct failures
make e2e-test-recommendation     # Recommendation service cache failure
make e2e-test-ad                 # Ad service failure — all requests fail
make e2e-test-all                # Run all 4 fault injection tests sequentially

Each test injects a fault into the otel-demo app via feature flags, then triggers an OpenSRE investigation to diagnose it.

EKS commands

Command	What it does
make e2e-setup-eks	Full setup: otel-demo on EKS + tunnels + agent + token
make e2e-teardown-eks	Uninstall otel-demo from EKS and stop tunnels
make e2e-status	Show cluster, pods, and observability status
make e2e-token	Generate a team token for web UI access
make eks-port-forward	Start port-forward tunnels to EKS
make eks-port-forward-stop	Stop port-forward tunnels

Local cluster (Kind)

For testing without a cloud cluster, you can use Kind instead:

make e2e-setup       # Create Kind cluster + install otel-demo + start agent
make e2e-teardown    # Delete Kind cluster and clean up

Comparing OpenSRE

OpenSRE — the memory-first AI SRE at opensre.in is not affiliated with Tracer’s OpenSRE at opensre.com.

• OpenSRE vs Tracer OpenSRE — team platform with episodic memory & knowledge graph
• vs commercial tools — PagerDuty AI, Rootly AI, Shoreline
• Blog: commercial comparison

Built With

OpenSRE is built on top of proven open-source technologies:

• Claude Agent SDK — Agent orchestration (root investigator → specialist subagents)
• Neo4j — Knowledge graph for service topology and dependency traversal
• FastAPI — Backend API with SSE streaming
• Next.js — Web console (dashboard, memory browser, config editor)
• LiteLLM — Optional multi-provider LLM proxy
• PostgreSQL — Config and agent-run storage (episodes live in Neo4j)

Star History

Contributing

We welcome contributions! See CONTRIBUTING.md for guidelines. Please open an issue before starting major work.

Contributors

Thanks to everyone who helps improve OpenSRE:

Creator

Swapnil Dahiphale · SRE · Builder "Built by someone who's been paged at 3am."

License

OpenSRE is licensed under the Apache License 2.0.