From raw signals
to running models.
One platform: call the models zero-shot through the API, or finetune Chronicle on your own data and deploy it as a private endpoint — with proof it works before you pay.
The API.
Send signal data and context, get answers. Metered per request, no platform fee — and OpenAI-compatible for embeddings.
Forecast→
Probabilistic forecasts for any signal, zero-shot. Add reasoning when text or image context should alter the result.
Embed→
Encode time-series, text, and metadata into one retrieval space for search, clustering, and classification.
Query your models→
Every model you finetune becomes an endpoint on the same API, called by name with the same keys and credits.
Get started in one call.
Choose forecast-only for fast probabilistic forecasts, or reasoning when text and image context should alter the result — including against your own finetuned forecaster.
Metered per request, finetuning priced on your data, GPU serving that costs nothing while idle — see pricing.
import httpx
response = httpx.post(
"https://inertialai.com/api/v1/forecasts",
headers={"Authorization": "Bearer iai_..."},
json={
"model": "inertialai-forecasting-reasoning",
"series": [18.2, 18.7, 19.4, 21.6, 24.8, 27.9],
"horizon": 6,
"context": "Promotion starts next week; expect a demand lift.",
"image_urls": ["https://example.com/promo-calendar.png"],
},
)
forecast = response.json()["final_forecast"]Finetuning, end to end.
Upload data, train Chronicle, see the proof, and ship a private endpoint — all from the dashboard or four API calls.
Upload your data
JSONL records mixing time-series and text. The shape of your records picks the objective — forecasting, classification, regression, text generation, or embeddings. Every file is fully validated, and priced, before you pay a cent.
Train with live curves
Training runs on isolated GPU compute with live loss curves in the dashboard. Priced on your data, from $5 a job.
Get proof, not vibes
Every job ends with a before/after eval against the frozen base on held-out data, exportable as a PDF report. If your model doesn't beat the base, the training fee refunds automatically.
Deploy a private endpoint
One call launches your model on the GPU you choose — metered per call, scales to zero, versioned with one-call rollback and hard spend caps.
Improvement guarantee
No improvement over the base model → automatic refund. You only pay for finetunes that work.
Scoped API keys
Mint keys that can only call one model — safe to embed in a device fleet or hand to a contractor.
Your data stays yours
Per-customer isolated storage, raw uploads deleted after training, hard-delete everything anytime. Never used to train our base models.
Ready for production teams.
For proprietary data, high-volume workloads, and regulated environments — including a guided 30-day pilot on your data.