Make it yours

You have now seen the decisions in a Physical AI data workflow: inspect recordings, define labels, review quality, select data, export it, and evaluate what changed. Apply those decisions to your own workspace through the documented interfaces.

Start with a read

The Python and TypeScript SDKs expose supported API operations for datasets, projects, tasks, exports, and other resources. Start by inspecting what your API key can access, then choose the operation your workflow needs.

pip install "avala[cli]"
avala datasets list --limit 10
avala exports list --limit 10

Set AVALA_API_KEY through your environment before authenticated commands. The quickstart explains how to obtain a key. Available operations depend on its permissions and your workspace configuration.

Connect an export to your code

For an existing project that has passed your review gate, set AVALA_PROJECT_UID to its UID. This example creates an export job and prints its identifier and current status; it does not train a model:

import os
from avala import Client

project_uid = os.environ["AVALA_PROJECT_UID"]
with Client() as client:
    export = client.exports.create(project=project_uid)
    print(export.uid, export.status)

Use the export wait or get operation to inspect completion before downloading. Adapt the output to your training loader and keep the export identifier with the training run.

Use a coding agent

Avala's MCP server connects supported API operations to coding tools. For a local stdio connection, provide AVALA_API_KEY in the process environment and follow the MCP setup guide.

The local stdio server exposes read operations by default. Its mutation catalog requires explicitly enabling AVALA_MCP_ENABLE_MUTATIONS; leave that setting off for initial exploration. Agent access still uses the permissions of the supplied credential.

Add event-driven handoffs deliberately

Where supported, webhooks can notify your systems about workflow events. Choose the events your integration actually handles, verify delivery, and make retries safe. An event notification does not replace a review or deployment approval.

Continue with your own data

• Quickstart — create a key and make a first request.
• Python SDK and TypeScript SDK — inspect supported resources and operations.
• CLI reference — run documented commands from your terminal.
• MCP setup — connect your coding tools.
• Book a demo — discuss your sensors, labeling needs, and training-data workflow.

Avala is Fleet Learning Infrastructure for Physical AGI. The next step is to choose a real data problem, define a reviewable output, and connect the parts of the workflow that your team needs.