Create, run, monitor, and manage GPU-powered experiments with automatic logging, QC gates, and status lifecycle.
Experiments API
Shipped today: GET /v1/experiments, POST /v1/experiments, plus the sub-collections /experiments/costs, /experiments/queues, /experiments/templates (GET + POST each). Per-experiment routes (GET /v1/experiments/{id}, /logs, /metrics) are planned, not yet live. Stream logs via the experiment-detail view in the web app or the Convex experiments.getLogs query.
Experiments are the core research unit in Hubify. Each experiment runs on GPU compute (or CPU), streams logs in real time, passes through QC gates, and stores results back to the lab.
Experiment Status Lifecycle
queued → running → pass / fail / blocked
Status
Description
queued
Waiting for a compute pod or manual trigger
running
Actively executing on a pod
pass
Completed successfully and passed QC
fail
Completed with errors or failed QC
blocked
Waiting on a dependency or manual unblock
Create an Experiment
string
required
Convex ID of the lab this experiment belongs to.
string
required
Human-readable experiment title (e.g., “MCMC Chain, Planck 2018 + BAO”).
string
Convex ID of the parent project, if applicable.
string
Custom identifier (e.g., “EXP-054”). Auto-generated if omitted.
string
default:"CPU"
Compute target: H200, H100, A100, or CPU.
object
Experiment configuration as a JSON object. Stored as a string internally.
Show Config fields
string
GPU type override.
number
Batch size for data loading.
number
Number of training epochs.
string
Script path to execute on the pod.
object
Environment variables to set.
string
Experiment template name (e.g., mcmc, anomaly-sweep, fisher-forecast).
string
default:"normal"
Queue priority: critical, high, normal, low.
number
Estimated runtime in seconds. Used for scheduling and cost projection.