Tsuga привлекла $35 млн на наблюдаемость AI-агентов без платы за каждый байт

Paris-based startup Tsuga raised $35 million in a Series A round — just six months after exiting stealth mode. The company is building an observability platform specifically for the era of AI agents, where telemetry volumes are growing faster than existing monitoring tools can handle.

Who's Behind Tsuga

The startup was founded by two former Datadog employees — the company that essentially created the modern observability market and is now valued at around $40 billion. Datadog alumni in the industry are a signal: people know the market from the inside and see an uncovered niche within it. Tsuga exited stealth mode less than a year ago and has already raised a serious round. This suggests that investors believe the niche is real, the timing is right, and the company needs capital now to capture the window.

Why the Old Model Is Breaking

Standard billing in observability is pay-per-byte of telemetry. For traditional web applications, this worked: a service generates predictable volumes of logs and metrics, pricing scales linearly. AI agents break this math. An agent orchestrator in a single session creates hundreds of nested tool calls, prompts, model responses, and traces — each needed for debugging, but together they generate volumes where per-byte billing becomes financially unbearable.

• AI agents produce orders of magnitude more telemetry than regular services
• Per-byte billing makes comprehensive monitoring impractical
• Data sampling hides critical errors in call chains
• Existing tools are not optimized for AI tracing patterns
• Companies are forced to choose between data completeness and budget

Tsuga wants to offer a different economics — not per-byte, but some other metric that makes comprehensive monitoring accessible even with explosive load growth.

Inside the Client's Cloud

The second principled choice of the startup is the deployment model. Tsuga is installed within the client's own infrastructure, rather than pulling telemetry data to external servers. For the enterprise market, this is critical. AI agent traces can contain prompt content, fragments of processed documents, user data. Handing all of this to a third-party vendor means creating legal and regulatory risks, especially in finance, healthcare, and government. The "in your cloud" model is not just an architectural choice, it's an argument in deals with large enterprises.

What This Means

The observability market for AI agents is just forming, and the first serious capital has already begun to flow. Founders with Datadog experience, rejection of the per-byte trap, and deployment in the client's infrastructure — Tsuga enters with a thoughtful set of arguments. If AI agents become mass infrastructure in the next two to three years, their monitoring tools will become a mandatory layer of the stack — and the race for this market is already on.