01 / PRESERVE
Keep the source in sight.
Every proposed value points back to its source. Keep the evidence alongside the interpretation.
Explore evidence
OPEN-SOURCE TYPESCRIPT FRAMEWORK
A data model forwhat comes next.
Ontave turns unexpected messages into useful records. Preserve the source, review the changes, and let your model grow without losing context.
01Source preserved
02Change reviewed
03Contract pinned
NOON, REST FROM WORK · VINCENT VAN GOGHA LITTLE ROOM FOR WHAT COMES NEXT
YOUR SCHEMA DIDN’T SEE THIS COMING.The data model your
The data model your
information actually needs.
New information doesn’t always fit the model you started with. Ontave preserves the evidence, proposes what’s missing, and keeps you in control of what changes.
02 / REVIEW
Every change. Your call.
Inspect a proposal before it becomes accepted data. Model approval and fact acceptance stay separate.
Understand review03 / EVOLVE
Room for the next concept.
Evaluate an additive model change before activating it. Keep existing application contracts pinned.
See model evolution04 / READ
Read data. Skip the inference.
Query accepted records through a typed contract. Your database answers without another model call.
Explore structured readsHOW ONTAVE WORKSOne message.
One message.
A new concept.
A familiar installation. An unexpected inspection. The source stays intact.
RameshSource 014
Current modelv1
○ Person
□ Installation
?
Where does an inspection fit?
Scroll to follow the story ↓
Try this example yourself
Illustrative walkthrough
Keep the context.
Build what comes next.
Explore the API ↗↳
Keep the source in sight.
Preserve the original message. Keep every proposed value connected to the evidence behind it.
Explore evidence ↗ Receipt ≠ accepted fact
const receipt = await ontave.ingest(context, {
source: "messages",
idempotencyKey: "message-014",
payload: {
kind: "text",
text: "Ramesh needs to inspect the roof.",
},
});
// The source is durable. Interpretation comes next.
const job = await ontave.jobs.get(
context, receipt.job_id,
);
CLEAR BOUNDARIES. BETTER FOUNDATIONS.Built on evidence.
Built on evidence.
Clear about uncertainty.
Fewer repeat calls.
More context kept.
How we measured it ↗80%Fewer extraction calls
Fewer extraction calls
on repeated deliveries.
100 sources. Five deliveries each. One extraction per source.
Compared with extracting every delivery.0 callsYour accepted data.
Short synthetic Person records; page size 25.
Your accepted data.
No inference on read.
1,000 structured queries. Zero model calls. Local read timings at three dataset sizes.
Read p95, 100 timed reads per dataset.Short synthetic Person records; page size 25.
100/100Accepted field assertions.
It does not establish semantic accuracy.
Accepted field assertions.
Every source attached.
Every accepted field assertion kept its source. None appeared before approval.
Source coverage measures traceability.It does not establish semantic accuracy.
Local alpha.2 measurements. Human time savings and provider cost savings are unmeasured.View the data ↗
Inspect the numbers and environment
Apple M4 Pro · 24 GB · Node.js 24.22.0 · localhost PostgreSQL 18.4
| Measurement | Result |
|---|---|
| Deliveries / distinct sources / extraction calls | 500 / 100 / 100 |
| Model calls for 1,000 structured queries | 0 |
| Source-referenced / accepted field assertions | 100 / 100 |
| Ordered-read p95: 100 / 1,000 / 10,000 records | 0.95 / 1.37 / 8.27 ms |
Synthetic fictional data and automated review. No provider calls. Not a production SLO, model accuracy score, billing comparison or human productivity study. Read the full method →
One message.
See where it takes you.
A local SDK session. One command.
Node.js 24. No Docker. No API key.
npx --yes --package=https://github.com/smileplis/ontave/releases/download/v0.1.0-alpha.2/ontave-0.1.0-alpha.2.tgz ontave playground --installA FEW THINGS TO KNOWA few things,
A few things,
made clear.
Is Ontave a database?
Ontave is an open-source TypeScript framework that runs on your server and uses your PostgreSQL database. It connects preserved evidence, reviewed assertions, and an evolving logical model—so your application can use information it didn’t anticipate.
What does the browser playground run?
The browser playground lets you explore fictional reference runs from the real SDK and PostgreSQL. Run the one-command local playground for live SDK and database operations. Both use a labelled fixture adapter, with no model calls.
Can I use my own messages?
Yes, through the SDK on your own server, with your database and a model adapter you choose. The guided playground uses supplied fictional scenarios. The Node.js integration guide shows how to connect your own data.
Does a new concept build my app for me?
Ontave makes room for a new concept in a reviewed data model. You build the screens and workflows around it. The generic playground can render a new contract dynamically; custom application interfaces still require development.
What can I install today?
Start with the developer preview: a versioned GitHub archive you can install with npm. It isn’t on the npm registry yet. The release status explains what’s available and what is still being qualified.
A LITTLE ROOM FOR WHAT COMES NEXTLet your information
GREEN WHEAT FIELDS, AUVERS · VINCENT VAN GOGHLet your information
find its place.
Start with one unexpected message. Keep the context.
Decide what comes next.