Run time series foundation models locally

Pull a model, run a forecast. No cloud API, no setup.

curl -fsSL https://www.routeframe.com/install.sh | bash
macOS Linux x86 Linux ARM Windows
# Your CPU has been climbing all week. What happens next?

$ routeframe run toto --input "45,48,52,49,55,58,62,59,64,67" --horizon 4

Predictions:
t+1: 68.42
t+2: 71.05
t+3: 69.88
t+4: 72.31

# Scale up before Thursday, not after the page.

Just 3 steps needed

1
curl -fsSL https://www.routeframe.com/install.sh | bash
15 MB binary. No dependencies.
2
routeframe pull toto
Downloads a foundation model trained on 2 trillion time series data points
3
routeframe run toto --input "45,48,52,49,55" --horizon 8
Predict the next 8 values. Runs on GPU automatically.

Forecasting use cases

Routeframe puts foundation model forecasting directly in your workflow - reduce infrastructure costs with proactive scaling, catch anomalies before they become incidents, predict demand before it spikes, and embed forecasting into any service through a local API.

Capacity planning

Scale before it hurts, not after

Predict CPU, memory, and traffic weeks ahead. Provision capacity proactively and cut cloud costs by eliminating emergency over-provisioning. Teach the model about deploy days with --exogenous is_deploy to account for known spikes.

Anomaly detection

Catch what's off before it pages

Routeframe forecasts what your metrics should look like. When reality diverges from the prediction, that's your signal. Pipe live metrics through routeframe monitor and catch anomalies in real time before they become incidents.

Demand forecasting

Know what's coming before it arrives

Forecast revenue, orders, API traffic, or inventory from a CSV with one command. Fine-tune on your own data to capture your business's seasonal patterns. Flag known events like holidays or launches with --exogenous is_holiday.

Embedded forecasting

Add forecasting to any service

Run routeframe serve and call POST /api/forecast from Go, Java, Python, or any language. 4ms latency. The model stays warm between requests. No ML dependencies in your service, no round-trip to a cloud API.

Why run it locally

Cloud ML APIs charge per prediction, add latency, and require sending your data off-network. Routeframe is a CLI that runs on your hardware - and because it exposes a local REST API, it can be used as a tool by any AI agent of your choice to fine-tune models like Toto on your data and reason over your time series forecasts.

Data stays on your machine

Your metrics never leave your network. No API keys, no data processing agreements, no compliance reviews.

No per-prediction costs

Run a million forecasts a day. It's your CPU. Cloud APIs charge $0.01-0.10 per call -- that adds up.

4ms, not 400ms

Local GPU inference is 100x faster than a round-trip to a cloud endpoint. Fast enough for real-time dashboards.

Works offline

Air-gapped environments, edge deployments, planes. The model runs with no internet connection.

What this replaces

BeforeWith Routeframe
Get a forecastAsk the data team, wait 2 weeksRun one command, get it now
Add to your serviceDeploy a Python ML servicePOST to localhost:11435
Train on your dataSet up PyTorch, write training looprouteframe finetune --data csv
Handle known eventsManual adjustments, tribal knowledge--exogenous holidays,deploys
DependenciesPython, PyTorch, CUDA, 4+ GB15 MB binary, nothing else

Available models

Toto

289 MB
by Datadog · 151M parameters · Apache 2.0

Foundation model trained on 2 trillion time series data points from real-world infrastructure, business, and IoT metrics. Handles seasonality, trends, and sudden shifts out of the box. Fine-tune it on your data to learn your specific patterns.

forecasting multivariate exogenous covariates fine-tunable
routeframe pull toto