Product

A pipeline that tailors, evaluates and deploys

Kethra is a fine-tuning, evaluation and deployment pipeline for companies that want a distinctive private model, without standing up an ML team. Bring your data and your standard. Keep the weights.

Start free How it works

01The pipeline

Fit, eval, deploy

Step 01

Fit

Fine-tune on your data and voice

We start from a strong open-weight base suited to your task, then fine-tune on your corpus and align the model to your written standard. Where it helps, we distil to smaller, faster weights.

kethra fit --data ./corpus --standard ./voice.md --base auto

Step 02

Eval

Score against a bar built from your standard

A held-out eval is built from your own examples: prompts, the responses you would accept, and the failures you would reject. Every version is scored on voice, accuracy and adherence. Nothing ships until it clears.

kethra eval --suite ./bar → pass 47/50 · voice +31

Step 03

Deploy

Ship to a private endpoint you own

One command puts the model behind a private, OpenAI-compatible endpoint, with versioning and rollback. Host on our infrastructure or inside your own. The weights are yours to export at any time.

kethra deploy --private → https://api.basisresearch.tech/v1

02What it does

Measured to you, part by part

Fine-tuning on your data and voice

fine-tune

We fit the model to the way your company actually writes and decides, from your own corpus, not a generic instruction set. The result reads like you on your best day.

Alignment to your standard

alignment

You write the standard once. Kethra aligns the model to it, so tone, refusals and house rules hold under pressure, not only in the demo.

Private, owned weights

weights

The tailored weights belong to you. Export them, host them, keep them behind your own walls. Owned, not rented from the same shelf as everyone else.

Evaluation against your bar

eval

Every version is scored against a held-out eval built from your examples. You see the delta over the base model before anything is deployed.

One-click deployment

deploy

Ship the model to a private endpoint in a single command, with versioning, rollback and an OpenAI-compatible API your engineers already know.

Ongoing refinement

refine

Standards move. Each cycle re-fits and re-evaluates the model on fresh material, so the fit stays true rather than drifting out of shape.

03The API

An endpoint your engineers already know

Your tailored model is served behind a private, OpenAI-compatible endpoint. Point your existing client at it, swap the base URL and the model name, and keep the code you already wrote. Versioning and rollback are built in, and the weights stay yours.

OpenAI-compatible chat and completions
Per-version endpoints with instant rollback
Private by default, your data trains your model only
Export the weights in a standard format, any time

Read the API quickstart

example.ts

import OpenAI from "openai";

const kethra = new OpenAI({
  baseURL: "https://api.basisresearch.tech/v1",
  apiKey: process.env.KETHRA_API_KEY,
});

const res = await kethra.chat.completions.create({
  model: "your-house-v3",   // your tailored model
  messages: [
    { role: "user", content: "Draft a renewal note." },
  ],
});

console.log(res.choices[0].message.content);

A model made to measure.

Start free and tailor a sample in a minute, or book a fitting and we will build you a private model of your own.

Start freeBook a fitting