Proposal

Agent-SRE is an AI-native SRE framework (1,071+ tests) that provides SLI/SLO tracking, chaos testing, canary deployments, and incident response for AI agent systems.

Integration Opportunity

OpenLit is OpenTelemetry-native with 50+ LLM provider integrations. Agent-SRE can complement this by adding the SRE discipline layer:

1. SLO-Based Alerting - Agent-SRE defines SLOs (latency P99 < 500ms, error rate < 1%) and tracks error budgets. These SLO metrics can be exported as OTel metrics that OpenLit visualizes alongside LLM traces.

2. Chaos Test Spans - Agent-SRE chaos experiments (fault injection, latency simulation, token exhaustion) emitted as OTel spans, visible in OpenLit's trace view alongside the LLM calls they affect.

3. Canary Metrics - During progressive rollouts of new LLM models/prompts, agent-sre tracks canary vs baseline. These comparison metrics enrich OpenLit's monitoring dashboard.

4. Error Budget Dashboard - OpenLit could display agent-sre's error budget consumption over time, helping teams decide when to deploy new features vs focus on reliability.

Proposed Approach

from agent_sre.exporters.otel import OTelSLOExporter
from opentelemetry import metrics

# Agent-SRE exports SLO metrics via OTel protocol
exporter = OTelSLOExporter(
    endpoint='http://localhost:4318',  # OpenLit OTLP endpoint
    service_name='my-ai-service'
)
slo_engine.add_exporter(exporter)

Since OpenLit already speaks OpenTelemetry natively, the integration is clean - agent-sre just needs to export its SLI measurements and chaos test data as OTel metrics and traces.

Our companion project Agent-Hypervisor also has a structured event bus that could export ring transitions, saga steps, and liability events via OTel.

Happy to submit a PR. See agent-sre for the full API.