GoodMem uses OpenTelemetry to trace retrieval and model calls in AI applications, making slow RAG responses easier to investigate in Langfuse, LangSmith, Jaeger, or Grafana Tempo.
In our evaluation of seven self-hosted memory systems, GoodMem received the highest overall observability score. Diagnostic detail and control over trace data were particular strengths.
GoodMem’s review · September 28, 2026
Scores reflect our assessment of the tested self-hosted software. They are not an independent certification or a speed benchmark. See versions and scope.
For a slow RAG response, retrieval tracing shows which stages ran and where the time went, including any LLM calls. The recorded outcomes also distinguish upstream failures from expected request rejections.
The OpenTelemetry (OTel) integration was tested with Langfuse and LangSmith for LLM observability, and with Jaeger and Grafana Tempo for infrastructure monitoring. In each case, the review verified trace delivery through the backend API.
Service metrics remain available through Prometheus. The documentation covers trace setup and the requirements for each backend, where access and retention are configured.
OpenTelemetry supportWith trace export enabled, GoodMem sends operational metadata and identifiers while excluding document content, prompts, generated answers, and credentials. Export is off by default.
Separate data policies govern application logs, retrieval records, and license reporting.
Enterprise securityThe assessment combined runtime tests with source and documentation inspection, focusing on enterprise deployments. Testing included failed requests and checks for sensitive test content in the exported traces.
The ratings reflect reviewer judgment across the six areas below, with the largest weights assigned to native OpenTelemetry support and diagnostic depth.
The comparison is dated September 28, 2026 and covers the versions below. Hosted services and separate commercial editions were excluded.
| Product | Version or commit |
|---|---|
| GoodMem | 95825e971 |
| Hindsight | v0.10.1 |
| Cognee | v1.6.1 |
| OpenViking | v0.4.22 |
| Graphiti | v0.30.2 |
| Supermemory | server-v0.0.8 |
| Mem0 | v2.2.1 |
PAIR Systems conducted the evaluation, testing GoodMem over several rounds with two separate scoring passes. Competitor reviews were shorter. All scores retain the judgments recorded at the time, including GoodMem’s known gaps.
Verification used synthetic workloads and backend APIs rather than dashboard inspection or production deployments. Privacy scores favor content exclusion by default or explicit opt-in capture.
We can walk through the evaluation with your platform team and discuss how GoodMem would fit your monitoring stack.
Talk to us