Sovereign Observability Needs Ownership, Not Just Residency
01

What is sovereign observability?

Sovereign observability is an operating model that keeps logs, metrics and traces inside infrastructure the organization controls, encrypted under its own keys, while still delivering the dashboards, alerts and investigations of a managed service. It differs from data residency: residency fixes where telemetry is stored, sovereignty fixes who controls it.

For a decade, sending telemetry to a vendor's cloud was a sensible trade: volumes were manageable and the vendor carried the operational burden.

On July 6, 2026, Tsuga published Bring your own agent, describing tooling that lets AI SRE agents query telemetry inside the customer's own cloud with the customer's approved model. Its premise: the intelligence layer on top of telemetry is becoming as sensitive as the telemetry itself.

On June 3, 2026, the European Commission proposed the Cloud and AI Development Act with four sovereignty assurance levels, warning that "over reliance on non-EU cloud service providers poses a significant risk to Europe's digital autonomy and resilience." As of November 2026, the proposal is still in the legislative process, and neither Parliament nor Council has adopted a position.

Residency is a location. Sovereignty is a control relationship, and observability data now falls squarely inside it.
02

Key takeaways

  1. Logs, traces and metrics routinely contain session tokens, personal data and LLM prompts, so telemetry is sensitive data.

  2. Data residency describes where telemetry is stored, while sovereignty describes who holds the keys, the storage and the right to leave.

  3. Cost-driven sampling and shorter retention thin out the evidence exactly when NIS2 and DORA reporting deadlines demand more of it.

  4. A bring-your-own-cloud model separates operating the platform from owning the data and removes the infrastructure margin built into traditional SaaS pricing.

  5. Organizations that own their telemetry can monitor every environment and run AI investigations inside their own perimeter.

03

Why is telemetry sensitive data?

Sovereign Observability Needs Ownership, Not Just Residency
Telemetry is sensitive because it records what systems did, including the data they handled. Application logs capture user identifiers, IP addresses, request payloads and stack traces, and AI workloads add prompts, retrieved documents and model responses.

The OWASP Logging Cheat Sheet lists session identifiers, access tokens, passwords, connection strings and sensitive personal data among items that should usually not be recorded directly in logs. Under incident pressure, those rules erode.

Where logs contain personal data, Regulation (EU) 2016/679 (GDPR) applies. Article 5 establishes "data minimisation" and storage limitation as core principles, the same logic that makes stale data an active risk rather than cheap storage.

Telemetry is not exhaust. It is a searchable record of your business and deserves the same classification as the systems it describes.

04

Is data residency enough for observability?

Not on its own. An EU region guarantees where telemetry is stored, but not who holds the encryption keys, who can access the data operationally, or what happens to it when the contract ends. Those three questions decide sovereignty.

This is not an argument against SaaS observability. For many teams, a managed platform in an EU region is the right choice: it removes operational burden, and contractual controls suffice for the data involved.

A third model has matured: bring-your-own-cloud (BYOC), where the vendor operates the software while the data plane runs in the customer's account. The three models compare as follows:
  • SaaS in an EU region: telemetry sits in the vendor's cloud account, the vendor usually holds the keys and operates the platform, infrastructure cost is bundled into the subscription with the vendor's margin, and leaving means exporting and re-ingesting data.
  • Self-hosted open source: telemetry, keys and operations all stay with the customer, and so does the engineering effort of running the platform.
  • BYOC, vendor-managed: telemetry sits in the customer's cloud account under its own KMS keys, the vendor operates the platform through a control plane, infrastructure is billed at the customer's own cloud rates without a markup, and the data stays in place on exit.
The question is not where your telemetry is stored. The better question is who can read it, who can switch it off, and what you keep when you leave.
05

Why does sampling become a compliance problem?

Sampling becomes a compliance problem when the data dropped to save money is the evidence needed after an incident. Regulators expect a fast, documented account.

Grafana Labs' 2026 Observability Survey, based on 1,363 responses collected between October 2025 and January 2026, found cost to be the most important tool selection criterion for 65% of respondents. Part of that cost is not software at all but resold infrastructure: according to Tsuga, traditional SaaS platforms add a 20 to 30 percent markup on the underlying compute and storage.

Directive (EU) 2022/2555 (NIS2) requires an early warning within 24 hours of becoming aware of a significant incident, a notification within 72 hours, and a final report within one month. Regulation (EU) 2022/2554 (DORA) sets comparable duties for financial entities and requires exit strategies for critical ICT third-party services.

Sampling is not a cost optimization. It is a decision about which evidence will not exist when you need it.

06

Why did Alan&Eve add Tsuga to its technology partners?

Alan&Eve added Tsuga because its architecture separates operating an observability platform from owning the data. The data plane runs inside the customer's AWS, Azure or Google Cloud account, telemetry sits in the customer's object storage under the customer's KMS keys, and Tsuga manages the software through a control plane connected via mutual TLS. Tsuga puts the model in six words: 

"Managed by Tsuga. Owned by you."

Tsuga

Portability

Tsuga is OpenTelemetry-native and needs no proprietary agent.

Pipeline-level redaction

A sensitive data scanner redacts personal data and credentials before storage.

Governance by default

SSO, role-based access, team boundaries, retention policies and audit logs are standard.

Agents inside the perimeter

The MCP server and CLI let teams run their own AI agents on complete telemetry within their boundary

A significant cost lever

Every observability platform needs compute and object storage, whichever provider runs it. In a traditional SaaS model those resources run in the vendor's account and reach the customer inside the subscription, with the vendor's margin on top. With Tsuga they run in the customer's own account at rates already negotiated with the cloud provider, and Tsuga charges one flat rate per gigabyte ingested for software and service, with retention included and no per-host fees.


That lever changes what teams can afford to see: staging, development and long retention windows no longer compete with the budget. Each account also includes a forward-deployed engineer who tunes pipelines and retention inside the customer's environment. Tsuga, founded in Paris in 2024, announced a $35 million Series A on June 23, 2026 and counts Le Monde, Camunda and Buk among its customers. As with any platform, the value comes when a strong product becomes best-in-use rather than merely best-in-class.

When the platform runs where the data already lives, sovereignty and cost efficiency reinforce each other.

07

How do you move toward sovereign observability?

Sovereign Observability Needs Ownership, Not Just Residency
The sequence works regardless of vendor and follows the order used when building observability that helps teams operate: visibility first, controls second.
  1. Map where telemetry goes today.

    List every agent, collector and integration that sends telemetry outside your cloud accounts, including staging and CI. The map shows reach. It does not yet tell you what is sensitive.

  2. Classify what telemetry contains.

    Sample real log lines, span attributes and agent traces from the busiest services and tag personal data, credentials, customer content and prompts. Classification turns a sovereignty debate into a concrete list. It does not replace fixing logging at the source.

  3. Draw the boundary by data class.

    Decide which classes must never leave infrastructure you control. Regulated and AI-related telemetry usually stays inside. The boundary does not decide the hosting model on its own.

  4. Standardize instrumentation on OpenTelemetry.

    Grafana Labs' 2026 survey found OpenTelemetry used for logs at 48% of organizations, with 37% citing freedom to switch vendors as a top reason. Vendor-neutral instrumentation keeps the boundary decision reversible.

  5. Redact at the pipeline, before storage.

    Mask and route telemetry at ingestion so sensitive fields never reach indexes or dashboards. Pipeline redaction catches what slips through source fixes. It does not clean data already stored elsewhere.

  6. Set retention and access per team.

    Define retention by data class and purpose, restrict access through SSO and roles, and give AI agents scoped, auditable access. This is where observability becomes governable. Cost ownership still needs per-team attribution.

  7. Plan the exit and test it.

    Document what you keep, in which format, and how long a switch would take. DORA expects exit strategies for critical ICT services, and the discipline helps everyone else. An untested exit plan is an assumption, not a plan. 

08

Sovereignty that slows incident response will be bypassed

Every sovereignty control fails if it makes the on-call engineer slower. When the secure path is a ticket queue, people export logs to a laptop or paste stack traces into a public chatbot.

The real test of a sovereign setup is how fast an engineer gets from alert to root cause at three in the morning. That requires one query interface across signals, fast search under incident load, pre-approved break-glass access with audit, and redaction that keeps debugging context.

If the secure path is slower than the shortcut, the shortcut becomes the process.

09

What leaders should measure

Track outcomes. Dashboard counts and gigabytes ingested are activity measures and say nothing about control.
  • Boundary coverage: the share of telemetry volume that stays inside infrastructure the organization controls.
  • Sensitive data leakage: sensitive data findings per week in stored telemetry.
  • Evidence readiness: time to produce a complete incident timeline, measured against the NIS2 72-hour window.
  • Cost per service: observability spend attributed to owning teams.
  • Exit time: estimated time to move platforms without losing retained data.
The board-level question is simple: how quickly can the organization explain an incident using evidence it owns?

10

What to review this week

  1. List every destination outside your cloud accounts that receives logs, metrics or traces.

  2. Search one day of logs from your three busiest services for tokens, email addresses and authorization headers.

  3. Check who holds the encryption keys for stored telemetry and whether you can revoke vendor access yourself.

  4. Identify which services or environments are sampled or unmonitored for cost reasons, and who decided.

  5. Rehearse NIS2: can your team write a 24-hour early warning from the data you actually retain?

  6. Ask your current provider in writing what happens to retained telemetry and dashboards if you leave.

The exercise usually reveals that the sovereignty question was answered years ago by whoever installed the first agent.
11

Frequently asked questions

What is the difference between data residency and data sovereignty?
Data residency specifies the geographic location where data is stored and processed, such as an EU region. Data sovereignty covers who controls the data: who holds the encryption keys, who can access it operationally, and whether the organization can leave with its data intact. Residency is one input into sovereignty.
Is BYOC observability more expensive than SaaS?
Usually not, and often the opposite. Every observability platform consumes compute and storage; in a SaaS model the vendor runs them and passes them on with a margin. With BYOC, the same resources appear on your own cloud bill at rates you already negotiated, and you pay the vendor only for software and service. The advantage grows with data volume and retention.
Does sovereign observability mean we have to self-host?
No. Self-hosting keeps control but places upgrades, scaling and resilience on your own engineers. Vendor-managed BYOC keeps the data plane in your cloud account while the provider operates the software. A shared, EU-hosted deployment with strict team separation can also be proportionate for smaller teams with less sensitive telemetry.
Do application logs contain personal data under GDPR?
Frequently, yes. IP addresses, user IDs, email addresses and request payloads commonly appear in application logs, and any of them can make a log line personal data under the GDPR. That brings purpose limitation, storage limitation and security obligations into scope. Pipeline redaction and purpose-based retention are the most direct controls.
How does NIS2 affect observability?
NIS2 turns observability into reporting infrastructure. Entities in scope must deliver an early warning within 24 hours of becoming aware of a significant incident and a fuller notification within 72 hours. Meeting those deadlines requires complete, retained and searchable telemetry, clear responder access, and confidence that evidence was not sampled away.
If telemetry stays in our cloud, is it secure?
Not automatically. Keeping telemetry in your own account removes a third party from the data path, but misconfigured buckets, broad IAM roles and unredacted secrets remain your problem. Sovereignty shifts responsibility toward you, so telemetry needs the same access reviews and key management as any sensitive data store.
12

Bring your telemetry back inside your boundary

Alan&Eve offers Tsuga in two deployment options. For smaller deployments, we run Tsuga in a shared AWS environment on European soil, with a private team per customer to protect confidentiality, a fast start with no infrastructure of your own to manage. For full deployments, we set up a full-featured, dedicated Tsuga tenant inside your own public cloud account, where data, keys and infrastructure cost stay entirely with you.

Either way, the engagement starts with a two-week onboarding sprint. In week one, we hold a working session with your platform and security leads to map where telemetry goes and classify what it contains. In week two, we connect two or three services through OpenTelemetry, activate redaction rules, and set retention and access per team. You receive the telemetry flow map, a running environment and a recommendation on which option fits your data; migrating every remaining service is not part of the sprint.

13

Sources

let's go