Skip to content
roar · GLaaS · TReqs

Do you actually know what happened in your last training run?

roar observes your training and runtime behavior without code changes. GLaaS stores the lineage as a source of truth. TReqs is a lightweight coordination layer on top. One CLI. No framework. No lock-in.

Install roar → See how it works
works with PyTorch, JAX, TF — no wrappers, no decorators
The reality

Most ML teams can't explain their own model.

Not because they're careless — because the tools that were supposed to track this are heavier than the work itself. So nothing tracks it. And then someone asks what changed.

01

You burned through $4k in credits last week. Nobody can say which run did it.

02

The best model on your leaderboard is three months old. You can't reproduce it.

03

Two people changed the config at the same time. Neither remembers what.

04

Your checkpoint is sitting in S3. You don't know what code produced it.

How it works

One command. No instrumentation.

Prefix your training script with roar run. A runtime observer attaches to the process and records what actually happened — not what a logger remembered to log.

  • env & Python deps
  • data versions
  • config & CLI args
  • GPU / CUDA state
  • metrics & stdout
  • model artifacts
  • git SHA & diff
  • cost & duration
install & run
$ curl -sSL get.treqs.ai | sh
$ roar login
→ linked to workspace: early-ml
 
# wrap any script — no code changes
$ roar run python train.py
$ roar run torchrun --nproc 8 train.py
$ roar run ./your-bash-pipeline.sh
 
$ roar runs
run_ab82f1 ✓ train.py eval 0.847 $12.40
run_9c0a3e ✓ train.py eval 0.826 $11.80
run_71f0d2 … train.py running $3.10
What you get

Outcomes, not dashboards.

The point isn't another UI to log into. It's being able to answer questions you couldn't answer before.

source of truth

Every run, recorded the same way. Not a convention — a file that exists.

reproducibility

Re-run any past run with the same env, data, and config. Not "close enough."

what changed

Diff two runs and see exactly what moved: code, data, hyperparams, hardware.

artifact trace

Every checkpoint points back to the exact run that produced it. No orphans.

cost attribution

Know which experiment cost what. Know which engineer spent what.

lightweight coordination

TReqs lets you queue and claim training without standing up an orchestrator.

TReqs

Just enough structure to train as a team.

TReqs is a training request: a small, shareable unit of "who wants to train what, with what, on which box." It's not a platform. It's not Kubeflow. It's a ticket, a claim, and a lineage link.

  • file a request: config + data + budget
  • any teammate (or you) can claim it
  • roar captures the run, GLaaS stores it
  • the request resolves with a real artifact
  • not an orchestrator
  • not a scheduler
  • not a platform play
  • not a new framework to learn

Use it when two people need to stop stepping on each other. Don't use it when you don't.

Pricing

Free for individuals. Paid when you're a team.

No "contact sales" for basic usage. No hidden tiers.

Free
$0 / forever

Individual use. The CLI and local lineage.

  • roar CLI, unlimited runs
  • local GLaaS store
  • run diff & replay
  • single workspace
Install roar
Team
$20 / seat · month

Shared lineage, API access, training requests.

  • everything in Free
  • shared GLaaS across the team
  • TReqs coordination layer
  • API & webhooks
  • cost & attribution reports
Start team trial
Enterprise
Custom

VPC, SSO, audit. When procurement asks.

  • everything in Team
  • self-hosted GLaaS
  • SSO / SAML, audit logs
  • support SLAs
Talk to us
Try it on your last run

Install roar. Point it at a script. See what you've been missing.

Thirty seconds to install. No account needed to start. If it's not useful on your first run, you haven't lost anything.

$ curl -sSL get.treqs.ai | sh
Install roar → Read the docs