Getting Started

Anveld is designed around a local-first workflow:

1. Install the SDK.
2. Create a project.
3. Iterate in simulation.
4. Train locally or remotely.
5. Package and deploy to edge devices.

The Fast Path

pip install anveld
anveld init warehouse-picker
cd warehouse-picker
anveld dev
anveld train --config training/ppo.yaml
anveld deploy --target jetson-fleet

The Core Loop

1. Initialize a Project

anveld init creates the project skeleton: configuration, robot definitions, training configs, deployment manifests, and tests.

2. Iterate in Simulation

Use anveld dev for local feedback with the configured simulator backend. The intended workflow is to make simulation the default environment for early development, not a separate late-stage track.

3. Train and Evaluate

Use anveld train for local training runs. Trainers are plugin-provided, so the SDK stays small while algorithms evolve independently.

4. Deploy

Use anveld deploy to package artifacts and target real hardware or fleets. Deployment material includes manifests, converted models, and runtime configuration.

Project Shape

Typical projects look like this:

my-robot-project/
├── anveld.yaml
├── robots/
├── policies/
├── training/
├── sim/
├── deploy/
└── tests/

Repository Contributor Setup

If you are working inside the Anveld monorepo rather than consuming the published packages, use the repo toolchain:

nix develop
uv sync --all-extras
moon run sdk:test
moon run cli:test

Where To Go Next

• Read SDK Reference for the protocol and data model.
• Read CLI Reference for the command surface.
• Read Training & Simulation for the platform workflow.