Kovati Docs

Integrations

Every provider and runner the family speaks to, what it needs from you, and what it costs.

Documented in depth

Providers are infrastructure, not the product. This page is the current list; the principle is on that page.

Motion

ProviderWhere it runsCostNeeds
Kimodo (NVIDIA)Docker on your GPU, another machine, or a rented oneFree locally · per hour rentedDocker, a Hugging Face token, and granted access to a gated Llama 3 model
UthanaHosted, your accountMetered — per generated or downloaded secondAn API key

Detail: MotionForge providers.

Speech

ProviderWhere it runsCostNeeds
ElevenLabsHosted, your accountMeteredAn API key
DeepL (translation)Hosted, your accountMeteredAn API key. Paid plugin
Pseudo (translation)LocallyFreeNothing — it exists so the pipeline is exercisable with no account

Detail: SpeechForge providers.

Face

ProviderWhere it runsCostNeeds
The engine's own solverIn-process, CPUFreeNothing. Models ship with UE 5.8
NVIDIA Audio2Face-3DA container on your GPU, or a rented oneFree beyond hardwareDocker. The lip-sync models are ungated; only the optional emotion detector is gated

Detail: FaceForge providers.

Mesh

ProviderWhere it runsCostNeeds
TRELLIS.2 (Microsoft)Docker on your GPU, or a rented oneFree beyond hardwareDocker, and granted access to a gated Meta model
MeshyHosted, your accountMetered, creditsAn API key
TripoHosted, your accountMetered, creditsAn API key

Detail: MeshForge providers.

Compute

What it is
Docker DesktopHow every local runner runs. The hub detects it, and can start it
RunpodRenting a GPU by the hour. Ranked by real per-datacenter stock against a price ceiling; weights persist on a network volume
Another machinePoint a project at a runner on a studio workstation. Its token lives in your machine's vault
Hugging FaceWhere gated model weights come from. One family-wide token

Engines and frameworks

Unreal Engine 5.8The only supported version today
BlenderOptional. The garment solver, the mesh round trip, and a parked motion normalisation pass
Narrative ProBridged by separate add-ons — see Narrative Pro add-ons
MetaHumanSupported as a face and body target. No MetaHuman asset is required by the face solver itself
Model Context ProtocolEvery core plugin ships a toolset

Two rules about keys

A gated model needs access, not just a token. Two routes here depend on a grant a human at the vendor has to approve, which can take days: the local motion provider's Llama 3, and the local mesh provider's DINOv3.

The motion provider measures whether your token can read its gated model and tells you. The mesh provider does not yet surface its grant on the Keys page — a known gap.

Account-level keys are shared across the family. A Runpod key and a Hugging Face token belong to you, not to a plugin, so they live at AutomationForge/Runpod and AutomationForge/HuggingFace and every plugin reads them. Set once.

See keys.

Adding your own

A provider registers a capability, declares its inputs, says whether it is metered, and reports its readiness. Core plugins never name a vendor, so adding a route is a plugin — not a change to anything above the provider line.

The contract is neutral; the options are the vendor's own, declared by the provider and shown as the vendor names them.

On this page