Summary

• OllyGarden announced $4 million in financing on 9 October.
• Minimum Viable Instrumentation allows Rose to propose baseline OpenTelemetry changes as reviewed pull requests.
• Its reported opportunity to remove up to 85% of logs is company evidence, not a universal achieved reduction.

Berlin observability company OllyGarden has raised $4 million and launched Minimum Viable Instrumentation, extending its Rose product from improving existing application telemetry to establishing a baseline where little or none has been implemented. The funding and feature were announced on 9 October, with the company positioning both around the growing use of AI tools to develop and operate software.

Investors in the round included Next Frontier Capital, Grand Ventures and ACTAI, while DIG Ventures, Datadog, Grafana Labs and Dash0 also participated. Their involvement is evidence of financing, not independent proof of product performance. OllyGarden says the new capability is available to existing enterprise customers and that prospective customers can obtain a free trial.

Applications generate logs, metrics and traces to help engineers understand errors and follow activity across services. Yet collecting more data can increase observability costs without showing what a customer-facing operation actually did. A trace may contain extensive database detail while omitting the identifier needed to connect the failure with a particular user or business transaction.

OllyGarden’s engineering account gives one concrete illustration from a customer environment: a batch job running for 59.7 seconds generated 633,718 spans, most from database activity, while none described the business operation being performed. The company says its redesigned plan would require six spans. These are its own case figures, not a controlled independent trial or an expected result for other systems.

Rose previously examined existing OpenTelemetry instrumentation and proposed improvements. Minimum Viable Instrumentation adds the ability to find code that lacks a usable baseline, inspect service boundaries and propose what should be recorded. The planned telemetry covers incoming requests, outgoing connections to services or databases, and selected business attributes that make an incident easier to investigate.

The implementation process described by OllyGarden begins with analysing the repository and writing an instrumentation plan. Rose then configures the OpenTelemetry software development kit, proposes targeted code changes and can adjust automatic instrumentation that produces duplicate information. The company says it tests locally where possible, using an OpenTelemetry Collector to inspect sample output before opening a pull request.

Engineers still decide whether to merge that pull request. This approval boundary is material because instrumenting an application can change what data it captures. Logs and traces may inadvertently contain credentials, personal information or sensitive business attributes. Reducing noise does not guarantee privacy; organisations must still decide which fields are necessary, how they are protected and how long they are retained.

OllyGarden reports finding opportunities to reduce log volumes by as much as 85% across customer work. Its engineering blog also reports duplication and trace waste in reviewed environments. These are vendor observations, and the announcement does not demonstrate that every customer achieves the same reduction or that cloud spending falls in direct proportion to the volume removed.

AI-assisted software development provides another reason to examine instrumentation early. When code is created rapidly across many repositories, engineering teams may struggle to review the logging and tracing introduced with each change. Automated operational agents also depend on the signals available to them; noisy or incomplete telemetry can complicate diagnosis even when an agent can process it quickly.

The engineering blog describes a separate demonstration involving an application with Apple and Google sign-in flows. By adding a provider attribute, Rose made it possible to distinguish the two paths in the resulting telemetry. That example establishes what the tool is designed to record, but not its accuracy across arbitrary applications. Similarly, local verification that a collector receives sample telemetry is a useful development check rather than proof that production incidents will always be diagnosed correctly. Both distinctions are important when assessing claims for AI-assisted operational engineering.

The commercial test is whether Rose can propose changes that engineers trust and maintain across varied languages and application architectures. Review effort, compatibility, incident investigation quality and actual observability charges are all measurable. With MVI available now, OllyGarden has moved beyond a planned feature; its wider operational benefits still depend on what customers can demonstrate in production.