Skip to content
Convoy
Sign inDiscuss your deployment
Contact

Edge and cloud inference for robots

Deploy AI models to real robots.

Convoy is building the inference layer between trained models and robot hardware: models served on the robot or in the cloud, tested in simulation, evaluated against a baseline, and recorded as they run.

Discuss your deployment

Platform

Inference, evals and data, in one workspace.

  • Jetson verified

    Edge and cloud inference

    Run a model on the robot’s own Jetson, or point a configuration at a model endpoint in the cloud. On the Jetson, every request leaves a trace with its tokens, latency and the release that served it.

  • In the workspace

    Simulation and evals

    Run MetaWorld tasks on MuJoCo with a learned SmolVLA policy beside a labeled scripted baseline. Each eval stays with the configuration it tested.

  • In the workspace

    Data capture

    Replay camera episodes from runs and evals in the same workspace as the device’s live telemetry and inference traces.

How it works

One request, through the chip.

A real prompt on a Jetson Orin Nano, from the network to the GPU and back, then the board’s power and temperature over a 15-minute soak.

Release

Package. Qualify. Release. One identity from bench to robot.

  • Package

    Building

    Bring the model, input and action processing, runtime and target configuration together as one identified release.

  • Qualify

    Building

    Check the release against task and runtime criteria, and keep the conditions and results with it.

  • Release

    Building

    Carry the identified configuration to the robot, with a basis for watching it and managing the next change.

  • Boundary

    By design

    Your controller and safety systems stay outside Convoy. Release evaluation is not a safety certification.

Status

What runs today, and what is still being built.

One text-inference configuration has physical verification, and the simulation path runs in the workspace. Robot motion control, fleet rollout and general robot-policy qualification are not shown as verified.

  • Simulation

    In the workspace

    The implemented simulation path.

    Benchmark
    MetaWorld on MuJoCo
    Learned policy
    SmolVLA
    Baseline
    Labeled scripted baseline
    Planner
    Qwen text planner, fixed skill
    Runs on
    Local CPU or an external endpoint
    Scope
    Simulation, not physical robots

  • Jetson Orin Nano

    Verified

    Physical text inference.

    Model
    Qwen2.5-1.5B-Instruct Q4_K_M GGUF
    Runtime
    llama.cpp with CUDA offload on 29/29 layers
    Processing
    Text-only conversation; 2,048-token context; up to 128 output tokens
    Hardware
    NVIDIA Jetson Orin Nano Developer Kit Super, 8 GB
    Controller
    No robot controller or motion-control interface in this demo
    Tested
    September 14, 2026

  • Robot release

    In development

    Where a design partnership starts.

    Discuss your deployment
    Package
    In development
    Qualify
    In development
    Release
    In development
    Fleet rollout
    Planned
    Safety
    Stays with your controller

Design partnership

Start with one deployment.

We’re speaking with robotics teams working on learned manipulation: picking, packing, sorting, assembly, and related tasks. Each potential partnership starts with a defined model, robot configuration, and deployment goal.

Discuss your deployment

A good fit usually has

  1. A trained policy

    a model that already performs the task in some setting

  2. A defined robot setup

    the arm, sensors, compute, and controller you deploy to

  3. A concrete deployment problem

    the part that is getting in the way today

FAQ

Questions about Convoy

Still unsure? Email founders@deployconvoy.com

What can I use today?

The Convoy workspace runs simulation evals and shows live telemetry and inference traces from a model running on a physical Jetson Orin Nano. The broader robot deployment workflow remains in development; design partnerships start with a defined model, robot configuration, and scope.

Will it work with any model or robot?

One text-inference configuration has physical verification: Qwen2.5-1.5B-Instruct Q4_K_M on our Jetson Orin Nano. This does not establish support for arbitrary models or robot motion. Each deployment still needs a defined processing pipeline, runtime, hardware, and controller configuration.

Does Convoy replace the robot’s controller or safety system?

No. Those responsibilities need explicit interfaces and boundaries. Release evaluation is not a substitute for the robot’s safety system or a safety certification.

Does inference have to run in the cloud?

No. The appropriate placement depends on the target system and task. We are not promising automatic movement between edge and cloud.

Does Convoy train the model?

Our initial focus is deployment: the execution pipeline and release workflow around a trained policy.

Is Convoy a robot manufacturer?

No. Convoy is building software infrastructure for deploying learned models onto robots.

Contact

Tell us about your next robot deployment.

Share your model, robot configuration, and the deployment challenge you’re working through.

Email us about your deploymentOpens your email app with the subject filled in.

Email

founders@deployconvoy.com

Helpful to include

  • Model: what it does and its interfaces
  • Robot configuration: arm, sensors, compute, controller
  • Deployment challenge: what blocks the next release

Please do not include proprietary model files, credentials, or sensitive operational data.