How uzair-23/agentic-sre works
Agentic-SRE is a synthetic-incident response product: an operator starts an investigation, the backend generates telemetry and runs an agent graph, and the operator reviews and approves or rejects a proposed fix. The approved path executes the action and produces a postmortem that adjusts future monitoring thresholds. The UI also exposes incident status and traces; a separate evaluation harness scores pipeline behavior against a golden set. The graph follows the README and sampled call sites; unsampled internal agent and UI wiring is not expanded.
Main entry points: graph.py Agent graph · api/main.py REST API · cli/main.py Operator CLI · harness.py Pipeline evaluation
Components 25
Operator interfaces
- Response dashboard
frontend/src/App.tsx - Incident views
frontend/src/components/IncidentDetail.tsx - Approval panel
frontend/src/components/ApprovalPanel.tsx - Trace viewer link
frontend/src/components/TraceViewerButton.tsx - Operator CLI
backend/cli/main.py
Incident service
- REST API
backend/api/main.py - Incident store
backend/api/main.py
Response pipeline
- Telemetry generator
backend/simulator/incident_generator.py - Agent graph
backend/agents/graph.py - Anomaly monitor
backend/agents/monitor.py - Root cause diagnosis
backend/agents/diagnosis.py - Fix proposal
backend/agents/remediation.py - Human approval gate
backend/agents/graph.py - Action executor
backend/agents/graph.py - Postmortem learning
backend/agents/postmortem.py - Input guardrails
backend/guardrails/injection_guard.py - PII scrubber
backend/guardrails/pii_scrubber.py - Risk classifier
backend/guardrails/risk_classifier.py
Incident state
- Monitor thresholds
backend/simulator/thresholds.json - Incident state schemas
backend/agents/schemas.py
Evaluation
- Pipeline evaluation
backend/evals/harness.py - Golden incidents
backend/simulator/golden_set.json
Other components
- SRE operator
- Langfuse tracing
- Groq inference
Connections 37
- SRE operator → Response dashboard: uses
- Response dashboard → Incident views: renders
- Incident views → Approval panel: shows gate
- Incident views → Trace viewer link: offers trace
- Response dashboard → REST API: requests incidents
- Approval panel → REST API: submits decision
- Trace viewer link → REST API: requests trace
- SRE operator → Operator CLI: runs investigation
- Operator CLI → Telemetry generator: generates logs
- Operator CLI → Agent graph: runs pipeline
- Operator CLI → Agent graph: resumes decision
- REST API → Incident store: reads and writes
- REST API → Telemetry generator: generates logs
- REST API → Agent graph: runs pipeline
- REST API → Agent graph: resumes decision
- Agent graph → PII scrubber: scrubs telemetry
- Agent graph → Anomaly monitor: checks anomalies
- Anomaly monitor → Monitor thresholds: reads thresholds
- Agent graph → Input guardrails: checks injection
- Agent graph → Root cause diagnosis: diagnoses incident
- Agent graph → Fix proposal: proposes fix
- Agent graph → Risk classifier: classifies risk
- Agent graph → Human approval gate: requests decision
- SRE operator → Human approval gate: approves or rejects
- Agent graph → Action executor: executes approved fix
- Agent graph → Postmortem learning: creates postmortem
- Postmortem learning → Monitor thresholds: updates thresholds
- Agent graph → Incident state schemas: uses state
- REST API → Langfuse tracing: records traces
- Agent graph → Langfuse tracing: records traces
- Anomaly monitor → Groq inference: infers
- Root cause diagnosis → Groq inference: infers
- Fix proposal → Groq inference: infers
- Postmortem learning → Groq inference: infers
- Pipeline evaluation → Golden incidents: loads cases
- Pipeline evaluation → Telemetry generator: generates logs
- Pipeline evaluation → Agent graph: evaluates pipeline