Experiment tracking that survives scale.

Recurvia is deep reinforcement learning lab. We built Metrana to log our foundation model training runs at full rate.

We recently opened our solution to other labs as partners: sustained 85k series entries and 5M+ data points per second with a bill that won't surprise you.

Book a Call Read the Benchmark

python

import metrana

metrana.init(
    workspace_name="my-workspace",
    project_name="my-project",
    run_name="quickstart-ml",
    config={"optimizer": "adam", "lr": 3e-4, "batch_size": 256},   # logged as run attributes
)

Metrana plugs into your training workflow with just a few lines of code — start capturing system-wide signals immediately.

Copied to clipboard

What's stable at step 1,000 can be the source of a collapse at 100,000.

Every tracker had a ceiling - rate limits, cardinality caps, sampling - and past it, logging less becomes the default. The signal that would have explained the failure is the one you were forced to drop.

The metric was there. It wasn't found fast enough.

Thousands of series, one dashboard. Finding the one that moved before the collapse means digging manually - by the time you have, the run's already burned the compute.

Forced under-logging hides the exact step that failed.

Environments, episodes, steps. A flat run_id can't hold that shape - so you're back to hacking together your own hierarchy on top of a flat table.

THE VISIBILITY GAP

Log everything your training systems produce. We keep up, structure it, and make it usable.

01 — Ingest everything

Full-rate logging of thousands of signals per run. No sampling, no dropping. Sustained 85k series entries and 5M+ data points per second.

02 — RL-native, not RL-only

Environments, episodes, and steps are first-class — not folder names.

03 — Transparent pricing.

Pay for what you log. Set a spending limit; we'll first slow you down instead of charge past it.

04 — Analysis agent (early access).

Explains what you're looking at today. Deep diagnosis is being built with design partners.

Built by a lab. Run as a partnership.

Metrana comes out of Recurvia (recurvia.ai), a reinforcement-learning lab. We built this infrastructure because our own training runs outgrew every tracker we tried - and we weren't willing to log less or slow our workflows to fit the tool.

Rather than keep it internal, we run it for partner labs: the same infrastructure, the same full-rate logging, maintained by a team that trains models and depends on it too. Partner labs get direct roadmap input - the RL features in the product exist because a lab needed them - and we publish our numbers, our methodology, and our limits.

Become a partner lab About Recurvia

1 / 4

Built to ingest, store, and query metrics at massive scale.

System-level visibility

Operate complex training systems with full visibility. Metrana structures thousands of signals into a coherent system view so nothing gets lost between components.

Built for multi-agent complexity

Track per-environment signals, rewards, and trajectories across every agent in your system. When behaviour emerges or breaks, you see exactly where and why.

Faster diagnosis and resolutions

Fix problems faster with clear, actionable recommendations. Metrana traces failures to their origin, not the symptom, so you know what started it and when.

Decisions grounded in data

Pinpoint root causes and take decisive action. Every recommendation comes from what the system is actually doing. Not heuristics, not guesswork.

Understand your training system. Optimise it with confidence.

Request Demo

Experiment tracking that survives scale.

Integrate in minutes

Existing trackers weren't built for this workload.