v3.2 · inference platform · GA

Deploy machine learning to your windowsill.

Repot is an API-first inference platform for houseplants. Point a webcam at your fern, pick a model, and find out what your fern already knew.

Deploy a plant Read the docs

● all systems operational 1 plant alive p99 4.2s cold start

Infrastructure for things that photosynthesize.

3 primitives, fully managed

F-01/registry

Versioned plant registry

Every photo of your monstera, tagged and immutable. Cut a release when it sprouts a leaf. Roll back to the version before you overwatered it. The photo is safe. The plant is a separate conversation.

F-02/infer

Sub-five-second cold starts

Your is-this-basil-dead endpoint sits at zero until you panic at 11pm. It spins up in 4.2 seconds, returns a verdict, and scales back down. Billed per leaf-second, rounded up.

F-03/autoscale

Autoscaling foliage

Buy a fourth pothos and we provision a fourth GPU. The plant does not know it has a GPU. The GPU does not know it has a plant. The invoice has been fully briefed.

From repotting to production in four calls.

quickstart.sh

Register the plant

Name it. We assign an endpoint, a region, and a billing identity it never asked for.

Point a webcam

Any RTSP feed aimed at the soil. We ingest at 30fps and discard 29 of them.

Pick a model

Pull a verified model from the registry, or fine-tune one on a plant you have already lost.

Run inference

Get a verdict, a confidence score, and a feeling. Scale to zero between panics.

# install
$ npm i -g @repot/cli
$ repot login

# register a plant
$ repot plants create greg \
    --species monstera \
    --region sill-north-1

# deploy a model and run it
$ repot deploy soil-moisture-vibes
$ repot run greg --model soil-moisture-vibes
  → "moist enough, emotionally and literally" (0.91)

Trusted by people who own too many plants.

customers / self-reported

I deployed a model to tell me the pothos needed water. It told me what I already knew, with 94% confidence. I have never felt so seen and so audited at the same time.

Dana ReyesSenior Plant Parent, two monsteras and one fiddle-leaf fig in hospice

We migrated the entire succulent shelf to Repot over a weekend. Three of them have since died, but the dashboards are immaculate and the latency is excellent.

Marcus LiuHead of Foliage Reliability, home office, no reports

The cold starts are faster than my last three therapists. I do not know what the model does. I run it every morning before coffee.

Priya AnandIndividual Contributor on a single snake plant

Pricing that scales with your foliage.

billed per leaf-second

Hobby

$0/mo

One plant, low expectations.

Shared GPU, when one is free
Cold starts only
No SLA on whether the plant is alive

Start free

Greenhouse ● popular

$290/mo

Dedicated compute, per plant.

One GPU pinned to each plant
Webcam ingestion at 30fps
Alerts your plant fully ignores

Provision compute

Arboretum

Let's talk

For people who call their apartment a collection.

Unlimited plants and regions
On-prem misting integration
A solutions engineer who has also killed a fiddle-leaf fig

Contact sales

live console · no signup

Run inference on a plant right now.

greg is online and waiting to be diagnosed.

open the demo →