Eve v0.1Evidence v0.2 · two-process Iroh · open source →

The language and infrastructure for physical AI.

At the center is Eve: an open-source language for AI systems, services, and accelerators to coordinate through one typed, verifiable conversation. Around it, Vantar builds the data, robotics, compiler, and deployment stack physical AI needs.

Explore EveEve on GitHubExplore Coldstack

Built for teams shipping physical AI — on the stack they already use

ROS 2MuJoCoNVIDIA IsaacIntel Loihi 2SpiNNaker 2BrainChip AkidaMCAPPyTorchIrohAutomergeMiren

Featured · Eve

Open source

A programming language for machines that coordinate machines.

Eve replaces separate client programs, server programs, and disconnected protocol definitions with one typed conversation graph. The compiler projects that agreement into compatible endpoint plans while preserving one inspectable identity across the data center.

One global program

The conversation is the computation.

Typed choices, streams, cancellation, and failures project into one checked local machine per role.

Authenticated data plane

Meaning survives the network.

Persistent Iroh identities and exact role/plan policy let independent servers execute the same identified Eve Wire.

Machine-authored change

Evolution stays governed.

Automerge sync runs as an authenticated Eve conversation, then passes through conflict, validation, and promotion gates.

58

Tests passing

4

Transports

2

Wire encodings

v0.1

Research prototype

Explore the Eve languageRead the source

Coldstack

Design partners

Robot data belongs on object storage.

Fleets generate 0.5–2 TB per robot per day. Coldstack keeps your raw MCAP in your own S3 bucket, builds a compact index, and answers one composable query — semantic, time-series, and metadata together. The retrieval layer for robot data.

Python
import coldstack

ns = coldstack.Client(api_key=...).namespace("fleet-a")

results = ns.query(
    text="gripper slipping on transparent object",   # visual semantic
    signal="torque_z > 5 for 2s and velocity < 0.1",  # time-series pattern
    filter={"robot_id": ["r-204", "r-207"]},          # metadata
    limit=50,
)

≥5×

Cheaper (target)

vs hot-storage stacks

1–3%

Index footprint

of raw log size

<100ms

Warm query

hot-namespace target

0 bytes

Raw data moved

stays in your bucket

Working engine, validated on real public robot logs — index footprint measured at 0.02% on a camera/lidar recording. In design-partner recruitment; not yet benchmarked at fleet scale.

Become a design partnerRead the design

Why Vantar

The tooling hasn't kept up. We're fixing that, layer by layer.

Robot data has no retrieval layer

Fleets generate 0.5–2 TB per robot per day. ~99.9% is never read again, but the failures and edge cases that matter need fast search across the entire corpus. Hot storage is ruinous, Glacier is unsearchable. Coldstack makes petabyte-scale robot logs searchable on object storage.

Robotics tooling is broken

Robotics developers spend 40–50% of their time on environment setup, dependency management, and build-system wrestling. Torq replaces the fragmented toolchain with one CLI — init, build, sim, deploy.

Neuromorphic has no DevTools

27+ SNN frameworks exist for simulation and training. Almost none exist for experiment tracking on neuromorphic hardware. Nuro fills the gap — record, track, visualize, and deploy across any chip.

Distributed AI has no shared language

Models, tools, memory, accelerators, and services are still stitched together with separate APIs and deployment files. Eve starts from one typed conversation and compiles compatible endpoint plans for every participant.

Products

Six layers, one mission.

Eve

Executable research · v0.1

The open-source, server-native language for typed and verifiable coordination between distributed AI systems.

Coldstack

Design partners

Vector, time-series, and metadata search over petabyte-scale robot logs.

Torq

Live

Build, simulate, and deploy robots with one Rust CLI. Replaces the fragmented ROS2 toolchain.

Nuro SDK

Live

Universal SNN compiler. Train on GPU, deploy to Loihi 2, SpiNNaker 2, or Akida. Zero code changes.

Vantar Cloud

Beta

Push experiments, compare runs across hardware, share recordings. Remote hardware access.

Dev Kit

Exploration

Hybrid event camera and neuromorphic processor. From sensor to inference at under 1mW.

Eve

Research · v0.1

The language servers speak to think together.

Eve begins with one typed conversation instead of separate client and server programs. Its compiler projects that graph into compatible endpoint plans, then preserves the same semantics across memory, TCP, authenticated QUIC, or Iroh. Automerge protects collaborative draft promotion, while Miren packages the current multi-node testbed.

Illustrative Eve
conversation Generate(prompt: Prompt) -> stream<Token> {
    roles gateway, router, expert[*]

    gateway -> router: prompt within 2ms

    choice router {
        cached { router -> gateway: CachedResult; end }
        infer(expert) {
            router -> expert: prompt
            expert -> gateway: stream<Token>
        }
    }
}

58

Tests passing

4

Transports

2

Wire encodings

v0.1

Research prototype

Explore EveView on GitHub

Torq

One CLI replaces 15 minutes of setup hell.

Terminal
$ torq init my-robot --template mobile
  Created my-robot/ — ROS2 Jazzy workspace ready

$ torq build
  Cached build — 3.2s (was 4 min with colcon)

$ torq sim
  Launching MuJoCo — warehouse.xml

$ torq deploy
  Deploying to jetson-orin@192.168.1.50 — running

75×

Faster builds

5

Templates

3

Simulators

1

CLI

View on GitHub

Nuro SDK

Train on GPU. Deploy to neuromorphic silicon. Zero code changes.

Python
import nuro

graph = nuro.Graph()
graph.add(nuro.neurons.LIF(128, tau=20e-3))

# Train on GPU — surrogate gradients
model = nuro.compile(graph, target="gpu", requires_grad=True)
model.fit(train_data, epochs=50)

# Same model, neuromorphic silicon
loihi = nuro.compile(graph, target="loihi")
spinn = nuro.compile(graph, target="spinnaker")

5

Backends

4

Neuron models

227

Tests passing

0

Code changes

View on GitHub

Roadmap

Data to CLI to language to compiler to cloud to silicon.

Now

Eve

Open source · executable v0.1 research

A graph-native language and runtime for server-to-server AI coordination. One typed conversation compiles into compatible endpoint plans across four transports.

Now

Coldstack

Design partners

Search over petabyte-scale robot logs, built natively on object storage. Raw MCAP stays in the customer's bucket; we host only the index. In design-partner recruitment.

Now

Torq CLI

Shipped · v0.1

Open-source Rust CLI for robotics. Scaffolding, cached builds, simulator orchestration, and SSH deployment. One tool replaces the ROS2 setup workflow.

Now

Nuro SDK + DevTools

Shipped · v0.8

Universal SNN compiler with experiment tracking. Train on GPU, deploy to Loihi 2, SpiNNaker 2, or Akida. Hardware-agnostic recording and visualization built in.

Next

Vantar Cloud

In progress · v0.9

Push experiments to the cloud. Compare runs across hardware. Share recordings. Remote access to neuromorphic chips and simulators you don't own yet.

Future

Vantar Dev Kit

Exploring

A complete edge AI module for event-based vision. Hybrid event camera paired with a neuromorphic processor. Nuro pre-installed.

Physical AI needs
a full-stack platform.

From petabyte-scale robot data to neuromorphic compilers to edge deployment — one platform.

Early access — Coldstack, Cloud & Hardware