agilerl-arena
agilerl-arena is the standalone Arena SDK + CLI package for AgileRL.
It provides:
- Python client for Arena workflows (auth, environment validation, experiment submission, deployment, inference)
arenaCLI for scripting and CI usage- The training manifest schema (
agilerl.arena.models): the Pydantic models every runtime trains from, plusarena manifest validateandarena manifest schema
This package is distributed independently from core agilerl, but exposes modules through the shared namespace:
from agilerl.arena import ArenaClient, Agent
from agilerl.arena.models import TrainingManifest
Core agilerl re-exports those models and adds only what a local run needs (environment construction, a buildable replay buffer).
Installation
Install the SDK and CLI on their own (no torch):
pip install agilerl-arena
Core agilerl depends on this package, so pip install agilerl includes it.
Quickstart
1) Authenticate
Preferred for CI/automation:
export ARENA_API_KEY="arena_pat_..."
Or interactive login:
arena login
2) Validate a training manifest
arena manifest validate path/to/manifest.yaml
arena manifest schema
3) Validate an environment
arena env validate --source path/to/my_env.py --name my-env
4) Check GPU memory (LLM manifests)
arena memory estimate path/to/manifest.yaml --gpu "NVIDIA L4"
# Longest context one dedicated L4 serving GPU can hold:
arena memory solve max_model_len --inference --gpu "NVIDIA L4" \
--model Qwen/Qwen2.5-7B-Instruct
estimate is a pre-submission gate (exit 0 fits, 3 over budget, 2 usage error).
solve holds every other input fixed and returns the largest value of one
field. Install agilerl-arena[hub] unless you pass --config path/to/config.json.
See agilerl/arena/memory/README.md.
5) Submit a training manifest
arena experiments submit path/to/manifest.yaml --project my-project
Python SDK example
from agilerl.arena import ArenaClient
from agilerl.arena.models import TrainingManifest
client = ArenaClient() # uses ARENA_API_KEY if set
TrainingManifest.get_validated("dqn.yaml")
client.validate_environment(
source="acrobot.py",
name="acrobot-env",
)
result = client.submit_experiment(
manifest="dqn.yaml",
resource_id="arena-medium",
project="my-project",
)
print(result)
Inference example
Inference uses the same credential as the rest of the CLI: a personal access token
from Profile then CLI API key, or the token arena login stores. Agent picks up
ARENA_API_KEY when you pass no api_key.
from agilerl.arena import Agent
agent = Agent("https://<deployment-id>.inference.agilerl.com", api_key="arena_pat_...")
action, _ = agent.get_action(observation)
Chat sessions
LLM deployments keep chat history, and the CLI keeps track of which conversation you are in. The first prompt starts one and later prompts carry on from it:
arena agent run my-chat-model
arena agent generate --prompt "My name is Sam."
arena agent generate --prompt "What is my name?"
Start a fresh conversation with --new-session, or end the current one with
arena agent sessions clear so the next prompt starts anew.
To move between conversations, arena agent sessions resume lists what is
stored and lets you pick one with the arrow keys:
arena agent sessions list # what is stored, current one marked
arena agent sessions resume # arrow keys, Enter to pick
arena agent sessions get <session-id> # read a transcript
arena agent sessions delete <session-id> # remove one for good
Who a session belongs to follows the deployment's memory scope, chosen on the
first deploy with arena agent deploy <experiment> --memory-scope user|organization.
It defaults to user and cannot be changed later.
Notes on packaging and imports
- Distribution name:
agilerl-arena - Python import namespace:
agilerl.arena - CLI command:
arena
agilerl-arena and agilerl intentionally share the agilerl.* namespace as separate packages.
5 maintainers
Metadata
5 maintainers
Release files for agilerl-arena 1.28.0
For a detailed explanation of source distributions (sdists) and built distributions (wheels), please see the package formats documentation.
Source distribution (sdist)
| File | Size | Uploaded | Provenance | |
|---|---|---|---|---|
| agilerl_arena-1.28.0.tar.gz | 201.1 kB | Without provenance | Details |
Built distribution (wheel)
| File | Interpreter | ABI | Platform | Reset |
|---|---|---|---|---|
| agilerl_arena-1.28.0-py3-none-any.whl | Python 3 | none | any | Details |
Total release size: 453.7 kB
Release files / agilerl_arena-1.28.0.tar.gz
| Download URL | agilerl_arena-1.28.0.tar.gz |
|---|---|
| Size | 201.1 kB |
| Tags | Source |
|
SHA-256 checksum How to use checksums |
6f8dd3540e01edcd8271cef4e34cf298f743b23e9dcfb090bf84d8d8375198f2
|
|
BLAKE2b-256 checksum How to use checksums |
38c0cb000de743d850246c95e7a3d10d22a2f44e4b38e218f3076b7cecb10269
|
| Upload date | |
|
Uploaded using Trusted Publishing? What is trusted publishing? |
Yes |
| Uploaded via |
uv/0.13.4 {"installer":{"name":"uv","version":"0.13.4","subcommand":["publish"]},"python":null,"implementation":{"name":null,"version":null},"distro":{"name":"Ubuntu","version":"24.04","id":"noble","libc":null},"system":{"name":null,"release":null},"cpu":null,"openssl_version":null,"setuptools_version":null,"rustc_version":null,"ci":true}
|
Release files / agilerl_arena-1.28.0-py3-none-any.whl
| Download URL | agilerl_arena-1.28.0-py3-none-any.whl |
|---|---|
| Size | 252.5 kB |
| Tags | Python 3 |
|
SHA-256 checksum How to use checksums |
b3caaab832bd8692b097f459639455ac34849d1e5d13399a6fbc6e364020b88b
|
|
BLAKE2b-256 checksum How to use checksums |
57f67595fc1434758b419cecfc2852f16db173c67dd0872f9b8d56c888035829
|
| Upload date | |
|
Uploaded using Trusted Publishing? What is trusted publishing? |
Yes |
| Uploaded via |
uv/0.13.4 {"installer":{"name":"uv","version":"0.13.4","subcommand":["publish"]},"python":null,"implementation":{"name":null,"version":null},"distro":{"name":"Ubuntu","version":"24.04","id":"noble","libc":null},"system":{"name":null,"release":null},"cpu":null,"openssl_version":null,"setuptools_version":null,"rustc_version":null,"ci":true}
|
Security for agilerl-arena 1.28.0
Provenance
Provenance describes where a file came from. On PyPI, provenance is shared via attestations, which provide a verifiable record of the build or publishing details. View details, limitations and caveats.
No provenance found
None of the files in this release were uploaded with provenance metadata.
A note on package security
PyPI is a repository for distributing packages, not a service for auditing them. While third-party security researchers and our staff do their best to detect and remove malicious packages, PyPI cannot guarantee the quality, security or suitability of any given package.
To learn more about protecting yourself from malicious actors, visit our package security documentation.
