bruceowenga/ulinzi-sre-agent

How bruceowenga/ulinzi-sre-agent works

Ulinzi monitors infrastructure metrics, turns detected alerts into shared incident state, classifies incidents with an LLM, and routes remediation by severity. Low/medium incidents use registered runbooks; high/critical incidents get an LLM-generated playbook and pause for operator approval. Both routes report to notification and observability sinks. The graph distinguishes monitoring, incident workflow, model/runbook execution, and reporting. Source excerpts show the main runtime calls; some monitoring and integration capabilities are documented in the README but their implementations were not sampled.

Main entry points: remediation.py Automatic remediation · main.py Runtime entry · graph.py Workflow graph · triage.py Triage agent

Components 22

Monitoring

Monitor agent agents/monitor.py

Incident workflow

Incident state state.py
Runtime entry main.py
SQLite checkpoint main.py
Workflow graph graph.py
Triage agent agents/triage.py
Configuration config.py
Prompt templates prompts

Response execution

LLM client agents/llm.py
Automatic remediation agents/remediation.py
Manual playbook agents/remediation.py
Runbook registry runbooks/__init__.py
Container recovery runbooks/container_oom.py
Disk cleanup runbooks/disk_pressure.py
Log rotation runbooks/log_flood.py

Reporting

Reporter agent agents/reporter.py

Other components

Prometheus
Model endpoint
Operator approval
ntfy
Loki
Grafana

Connections 29

Runtime entry → Monitor agent: starts main.py
Monitor agent → Prometheus: queries agents/monitor.py
Monitor agent → Runtime entry: emits incident main.py
Monitor agent → Incident state: creates agents/monitor.py
Runtime entry → Workflow graph: invokes main.py
Runtime entry → SQLite checkpoint: persists graph state main.py
Workflow graph → Triage agent: dispatches graph.py
Workflow graph → Automatic remediation: routes auto graph.py
Workflow graph → Manual playbook: routes manual graph.py
Workflow graph → Reporter agent: dispatches graph.py
Triage agent → LLM client: requests classification agents/triage.py
Automatic remediation → LLM client: requests playbook agents/remediation.py
LLM client → Model endpoint: calls configured endpoint agents/llm.py
Triage agent → Prompt templates: reads triage prompt agents/triage.py
Automatic remediation → Prompt templates: reads playbook prompt agents/remediation.py
Triage agent → Configuration: reads model settings agents/triage.py
Automatic remediation → Configuration: reads model settings agents/remediation.py
Automatic remediation → Runbook registry: looks up runbook agents/remediation.py
Runbook registry → Container recovery: registers runbooks/__init__.py
Runbook registry → Disk cleanup: registers runbooks/__init__.py
Runbook registry → Log rotation: registers runbooks/__init__.py
Manual playbook → Operator approval: requests decision agents/remediation.py
Runtime entry → Operator approval: collects decision main.py
Reporter agent → ntfy: pushes notification agents/reporter.py
Reporter agent → Loki: writes incident log agents/reporter.py
Reporter agent → Grafana: posts annotation agents/reporter.py
Triage agent → Incident state: updates classification agents/triage.py
Automatic remediation → Incident state: updates action agents/remediation.py
Reporter agent → Incident state: updates report agents/reporter.py

Drawn by GitDiagram on November 14, 2026 from the repository's file tree, README and source files. Read as Markdown ·