Managing experiments

Creating an experiment

hive create exp -c hive.yaml

The -c flag is required and points to the experiment config file. The experiment name is specified inside the YAML via the experiment_name field. Adding a - suffix to experiment_name appends a random unique ID — we recommend this to avoid name clashes. You can override config values without editing the file by passing key=value arguments:

hive create exp -c hive.yaml \
  experiment_name=foo- \
  repo.branch=feature/try-hive \
  sandbox.resources.cpu=1

Listing experiments

$ hive list exp
NAME                  AGENTS  STATUS          AGE
my-experiment-abc123  10/10   Running         5m
test-exp-xyz789       0/10    ImageBuilding   2h
debug-exp-abc789      5/10    InProgress      15m

Column	Meaning
NAME	Experiment identifier
AGENTS	`ready/total` — e.g. `8/10` means 8 of 10 agents are ready
STATUS	Current experiment phase (see below)
AGE	Time since creation

Experiments progress through these phases:

Status	Description
`Pending`	Experiment submitted, waiting to start
`ImageBuilding`	Building container images
`InProgress`	Agents are being initialized
`Running`	Agents are executing experiments
`Completed`	All agents finished successfully

Viewing experiment details

$ hive get exp my-experiment-abc123
apiversion: v1alpha1
name: my-experiment-abc123
spec
  repo
    source:             https://github.com/your-org/repo.git
    branch:             main
    evaluation_script:  evaluation.py
    target_code:        main.py
  runtime
    agents:          10/10
    max_runtime:     -1
    max_iterations:  -1
  sandbox
    base_image:          python:3.14-slim
    evaluation_timeout:  60s
    setup_script:
      pip install -r requirements.txt
    resources:
      cpu:    2
      memory: 4Gi
status
  created: 2026-04-22T10:30:00Z (2h ago)
  phase:   Running
  message: Experiment is running successfully

Output sections:

name: Experiment identifier
spec: Experiment specification
- repo: Source code repository configuration
- runtime: Execution settings (agents shows ready/total, e.g., 10/10)
- sandbox: Container environment and resources
status: Current state
- created: Creation timestamp and age
- phase: Current lifecycle phase
- message: Condition message (if available)

Streaming logs

Stream logs for a running experiment with hive logs. The --source flag selects where the logs come from:

# Follow the coordinator logs live
hive logs my-experiment-abc123 --source coordinator

To inspect a specific sandbox worker, use --source sandbox together with --worker (0-based):

hive logs my-experiment-abc123 --source sandbox --worker 0

By default logs are followed live. Pass --no-follow to print the existing logs and exit:

hive logs my-experiment-abc123 --source coordinator --no-follow

Stopping experiments

hive stop exp my-experiment-abc123

To stop multiple experiments at once:

hive stop exp exp-1 exp-2 exp-3

When stopping more than one experiment you’ll be asked to confirm. Add --yes (or -y) to skip the confirmation prompt:

hive stop exp exp-1 exp-2 exp-3 -y

Viewing the dashboard

The experiment dashboard, which contains detailed information about all Hive experiments launched by your team, can be accessed by using:

hive dashboard

Use --no-browser to print the URL without opening a browser.

Getting started

Using the Hive

Releases

Creating an experiment

Listing experiments

Viewing experiment details

Streaming logs

Stopping experiments

Viewing the dashboard

​Creating an experiment

​Listing experiments

​Viewing experiment details

​Streaming logs

​Stopping experiments

​Viewing the dashboard

Creating an experiment

Listing experiments

Viewing experiment details

Streaming logs

Stopping experiments

Viewing the dashboard