Build your own AI SRE agents. The open source toolkit for the AI era.
8.1k
Stars
1.1k
Forks
165
Open issues
30
Contributors
AI Analysis
OpenSRE is an open-source framework for building AI-powered Site Reliability Engineering agents that automate incident investigation and remediation. It integrates with 60+ observability and communication tools (Datadog, Grafana, Slack, etc.) to help SRE teams investigate production incidents across logs, metrics, and traces. This is a specialized tool for SRE and DevOps practitioners, not a general-purpose framework—it specifically targets organizations that need autonomous incident response...
Inferred from signals mentioned in the README (tests, CI, type safety) — not a review of the actual code.
AI's overall editorial judgment — not an average of the bars above, can weigh other factors too.
OpenSRE: Open-source AI agent framework for automated production incident investigation
OpenSRE is a Python framework for building AI-powered Site Reliability Engineering agents that investigate production incidents by connecting to 60+ observability tools (Kubernetes, CloudWatch, EC2, etc.). It targets platform and SRE teams who want self-hosted, customizable AI incident response without relying on proprietary SaaS. Its secondary goal is providing a reinforcement learning environment for training AI agents on synthetic infrastructure failure scenarios — analogous to SWE-bench for coding agents but applied to production debugging. Still in public pre-alpha (v0.1) as of evaluation date.
Created January 2026 by Tracer-Cloud, an organization already operating in the observability/tracing space. Sponsored by Greptile. Project reached public visibility rapidly in the first half of 2026, coinciding with growing market interest in AI agents for operations.
7,243 stars in roughly five months suggests a strong initial reception, likely driven by the novelty of applying the SWE-bench benchmark analogy to SRE, Reddit/Hacker News coverage, and the Trendshift badge indicating it trended on GitHub. However, the current 78 stars/week is moderate deceleration from what must have been a strong early spike. Growth appears organic with some community momentum rather than sustained virality.
Adoption not verified. No case studies, user testimonials, or documented production deployments appear in the available README excerpt. The project explicitly labels itself pre-alpha, making production adoption at scale unlikely at this stage. Community engagement exists via Discord.
Appears to be a CLI-first Python application (REPL interface via 'opensre' command) with a plugin/integration layer for connecting to infrastructure tools. Likely consists of an agent orchestration core, a tool-adapter layer for 60+ integrations, and a test harness with synthetic incident scenarios. The test structure (tests/synthetic, tests/e2e) is explicitly documented and separates local vs cloud-backed scenarios, suggesting deliberate architectural discipline. Installation via curl-to-bash, Homebrew, or PowerShell implies a compiled or bundled binary distribution alongside the Python core.
Partially documented in README. Synthetic RCA test suites with scoring (root-cause accuracy, required evidence, adversarial red herrings) and real-world end-to-end tests against Kubernetes, EC2, CloudWatch, Lambda, ECS Fargate, and Flink are mentioned explicitly. Coverage percentages and unit test counts are not documented in README.
Last push June 20, 2026 — one day before evaluation date — indicates active, ongoing development. CI badge is present. Pre-alpha status is honestly disclosed. Given the project is only ~5 months old and pushing nearly daily, maintenance posture appears strong for this stage.
ADOPT IF: you are an SRE or platform team comfortable with pre-alpha software, want to self-host AI-driven incident investigation, and are willing to contribute to or work around rough edges in exchange for early access and customizability. AVOID IF: you need production-grade stability, enterprise SLA, or reliable incident automation in critical systems — pre-alpha status makes this unsuitable for mission-critical use today. MONITOR IF: you are evaluating the AI-for-operations space and want to track whether the SWE-bench analogy for SRE proves out as a training and evaluation methodology, which would have significant downstream value regardless of this specific project's trajectory.
Independent dimensions
Mainstream potential
5/10
Technical importance
7/10
Adoption evidence
2/10
- Pre-alpha status means APIs, integrations, and workflows may change significantly without stability guarantees, creating adoption and migration risk for early users.
- The RL training environment concept is ambitious but unproven — there is no public evidence that synthetic incident simulations transfer meaningfully to real-world RCA quality improvements.
- Tracer-Cloud is a relatively new organization; long-term maintenance continuity and commercial model sustainability are uncertain given the project's early stage.
- The '60+ integrations' claim in a pre-alpha project may reflect aspirational scope rather than fully implemented, tested connectors — actual integration depth cannot be verified from README alone.
- Competing with well-funded commercial AIOps platforms requires significant ongoing investment; without strong community adoption or commercial backing beyond the Greptile sponsorship, the project may struggle to close the quality gap.
Likely to mature into a credible open-source option for self-hosted AI incident investigation by late 2026 if community momentum holds. The RL environment angle may attract research interest independent of the agent tooling.
Newsletter
Get analyses like this every Monday
Free weekly digest of the most interesting open-source discoveries.
Languages
Information
- Language
- Python
- License
- Apache-2.0
- Last updated
- 7h ago
- Created
- 6mo ago
- Analyzed with
- anthropic/claude-haiku-4-5
Stars over time
Contributors over time
Top 100 contributors only — repos with more will plateau at 100.
Open issues
[BUG] Streaming investigation silently swallows pipeline errors when a deferred stage import fails
[BUG] Interactive shell masks real REPL errors as "generator didn't stop after throw()" when alert listener is active
[BUG] Parallel tool calls emit no trace spans — ThreadPoolExecutor workers lose the session-trace contextvar
[BUG] Every trace_span re-reads the whole session JSONL (O(n^2) per session) and pollutes the message parent-chain
[IMPROVEMENT] Test all integrations: verify OpenSRE provides complete setup instructions (including out-of-tool steps)
Top contributors
Similar repos
Arize-ai/openinference
OpenInference is a set of OpenTelemetry conventions and instrumentation plugins...
dastergon/awesome-sre
Awesome SRE is a curated reference list of Site Reliability Engineering and...
| Repository | Stars | Week Δ | Language | Score | Updated |
|---|---|---|---|---|---|
|
Tracer-Cloud/opensre
this
|
8.1k | +336 | Python | 7/10 | 7h ago |
 SmythOS/sre
|
1.3k | — | TypeScript | 7/10 | 3mo ago |
|
Arize-ai/openinference
|
1.1k | — | Python | 7/10 | 17h ago |
|
dastergon/awesome-sre
|
13.3k | — | — | 7/10 | 11mo ago |
 beenuar/AiSOC
|
1.5k | — | Python | 8/10 | 4d ago |
 superloglabs/superlog
|
1k | — | TypeScript | 7/10 | 18h ago |
Commercial SaaS with mature integrations and enterprise support. OpenSRE offers self-hosted control, open-source customizability, and no vendor lock-in, but lacks PagerDuty's production maturity and reliability guarantees.
Grafana targets incident management and alerting workflows with broad observability stack integration. OpenSRE focuses specifically on AI agent-driven root cause analysis rather than incident coordination, occupying a narrower but distinct problem space.
These are incident management platforms focused on process coordination. OpenSRE is focused on automated technical investigation, making them potentially complementary rather than directly competing.
Shoreline offers automation runbooks and remediation; Blameless focuses on SRE workflows and SLOs. OpenSRE's distinguishing angle is the AI agent + RL training environment combination, which neither competitor currently emphasizes.
A curated resource list, not a framework. Not a competitor but contextually relevant as the most-starred SRE repository. OpenSRE targets a fundamentally different use case — active tooling vs passive reference material.
Grafana (Loki · Mimir · Tempo · annotations) · 
Datadog · Honeycomb · Coralogix · 
CloudWatch · 
Sentry · Elasticsearch · Better Stack · Splunk · Victoria Logs · SignOz · OpenObserve · OpenSearch · Azure Monitor · Hermes
Kubernetes · 
AWS (S3 · Lambda · EKS · EC2 · CloudTrail · Bedrock) · 
GCP · 
Azure · ArgoCD · Helm · Jenkins