opensre vs superlog
Open-source framework for AI SRE agents plus the RL training and evaluation environment they need — connect 60+ tools you already run and investigate incidents on your own infra. — versus — Agentic observability for OpenTelemetry: ingest traces, logs and metrics, group noisy signals into incidents, then let pluggable agent runners investigate while you sleep. Self-hosted, open-core.
Both put agents on incident duty. OpenSRE is a framework plus RL environment for building and evaluating SRE agents against 60+ tools you already run; Superlog is the observability product — ingest, fingerprint, group — with agent runners hanging off it. Build versus operate.
| opensre | superlog | |
|---|---|---|
| Stars | 11k | 1.4k |
| Forks | 1.5k | 103 |
| Language | Python | TypeScript |
| License | Apache-2.0 | Apache-2.0 |
| Last activity | 3 days ago | 2 days ago |
| Topics | agents, evals | agents |
| Curated connections | 3 | 2 |
opensre — the curator's take
The most serious open attempt at 'SWE-bench for production incidents': not just an agent that greps logs, but a reinforcement-learning environment where SRE agents can actually be trained and scored on distributed failures. Connect the observability stack you already run (60+ integrations) and investigate on your own infrastructure. NOT production-stable — it's a public alpha and says so, APIs will move; and mind the telemetry section plus the vendor (Tracer Cloud) gravity. Adopt the benchmark and ideas today, bet the pager on it later.
superlog — the curator's take
Worth a look if you already emit OTel and want triage automated: it fingerprints and groups signals into incidents, then hands them to an agent runner you install into your coding agent through their skills pack, instead of paging a human first. The community edition here is genuinely usable and self-hosted — web app, API, OTLP proxy, workers, ClickHouse-backed queries, Postgres schema. Read the open-core line carefully: the shipped default runner only records a local incident summary, so the investigation runtime is yours to plug in or theirs to host, and 'self-heal' is the pitch rather than something the free edition does unattended. YC-backed and early; needs Postgres, ClickHouse and Docker.