Self-hosted traces, logs and metrics that you and your agents can query
Emerged finds new products worth a look, daily — what’s new today
Apgar
Compares each agent release with the last one using live production traces
Is this yours? Claim itIf you change prompts or tools on a production agent, this tells you within hours whether the new release loops, errs or costs more than the one before, without paying a judge model.
Apgar reads traces from Langfuse or any OpenTelemetry source and, for the first hours after a deploy, scores the new release against the previous one on eleven signals such as tool errors, retries, loops, answer length, latency and cost per turn. Regressions arrive as one Slack message or webhook naming where they cluster. It stores no prompt or completion text and admits it cannot spot a wrong answer that otherwise looks normal. Self-host with Docker Compose or Helm alongside ClickHouse.
More like this
Inventory every AI agent on your machine and what it can reach
A public record of when hosted language models quietly change behaviour
Run dozens of agent evals per trace in one sub-second request, no judge prompt
A local ledger of every tool call your coding agents make
Follow the models, agents and labs you depend on and hear when something breaks