BYOK, WITHOUT THE UPKEEP
Add BYOK with a model list and one simple call.
Keep your model choices fresh: update the catalog and ship a release. No model-list code
to rewrite.
TypeScript · JavaScript · Python / MIT licensed
import * as byokit from
'byokit/catalog';
const key = process.env.DEEPSEEK_API_KEY;
if (!key) throw new Error('Set DEEPSEEK_API_KEY');
const [model] = byokit.getModels(byokit.DEFAULT, {
providerIds: byokit.exact(byokit.ProviderId.DeepSeek),
sort: byokit.ModelSort.PriceAscend,
limit: 1, local: false,
});
if (!model?.connection) throw new Error('No model');
const result = await byokit.call(
model, 'Hello', key,
);
console.log(result.choices[0].message.content);01 / KEY HIGHLIGHTS
Set it up once.
Let automated catalog updates and verified tooling handle the
rest.
One simple call to get an up-to-date list of BYOK models, and one simple call to run inference. No provider-specific SDK glue required.
Curated presets paired with composable matchers (exact,
any, contain, exclude, and,
or, xor, not) for modalities, required
speciality, model tiers, context, and tag exclusions. Existing array filters remain
supported.
Automated 12-hour sync and scheduled releases keep catalog choices current, backed by reproducible community inference verification and badges.
Built-in calls keep keys in memory and send them directly to the selected endpoint. Custom storage and transport integrations remain under your app’s control.
Generate a smaller static model list for fixed queries. Dynamic queries use a bundled snapshot and deterministic in-memory caching.
Validate token budgets, context limits, features, and modalities before sending tokens or incurring API costs.
Full parity across JavaScript/TypeScript and Python with shared schemas, enums, query outcomes, and budget checking.
02 / PLAYGROUND
Edit a query. Press Run.
See exactly what byokit returns.
Your console output will appear here.
Ready.
Runs locally in an isolated sandbox. No API key needed, no provider requests, and no saved code. Supports offline byokit APIs; imports and network access are unavailable. Try a real model call →
03 / TRY IT
These results come from byokit itself.
Change the list, add a filter, inspect the
return.
| Provider / model | Context | USD / 1M tokens | Evidence |
|---|
DEFAULT shows reviewed recommendations. COMPLETE includes every upstream provider and output modality, including retired and uncertain offerings. Catalog presence does not verify availability or account access. Prices shown are USD per million tokens; media pricing may be unknown.
04 / HOW TO CALL THE API
Select an offering, provide that provider’s key, and call it. The direct helper reads its endpoint and protocol; reusable clients need both fields explicitly.
Catalog queries run offline. No API key or provider request is needed.
Full instructions in README ↗import * as byokit from 'byokit';
const key = process.env.DEEPSEEK_API_KEY;
if (!key) throw new Error('Set DEEPSEEK_API_KEY');
const [model] = byokit.getModels(byokit.DEFAULT, {
providerIds: [byokit.ProviderId.DeepSeek],
local: false,
limit: 1,
filters: { callable: true },
});
if (!model?.connection) throw new Error('No matching callable model');
const response = await byokit.call(model, 'Hello!', key);
console.log(response.choices[0].message.content);An array of provider/model metadata.
Install now: npm install byokit · pip install byokit
INTEGRATION CONTRACT
Copy a runnable example, then preserve these fields when wiring it into your app.
Pass the entire model to byokit.call(model, prompt, key). With a BYOKit
client, copy connection.baseUrl and connection.protocol,
then send apiModelId as the request’s model string. The offering’s
id is for selection and records.
DEFAULT provides maintained recommendations.
COMPLETE includes unsupported, deprecated, and media offerings.
callSupport.textGeneration means adapter compatibility; account access,
quota, lifecycle, and verification are separate.
Enter URL, protocol, and the exact served model ID, then choose
Use this model. The returned Local entry is ready for
call(). Finding models and verifying are explicit actions.
Image/audio/video generation and streaming execution are not supported.
Use the selected providerId’s key, including for resellers. Built-in
calls keep it in memory and require HTTPS for authenticated remote servers; loopback
HTTP supports local development. Custom storage adapters can persist keys. Python
clients block credential serialization.
Complete integration contract and naming reference in README ↗
05 / PRODUCTION OPTIMIZATION
Pre-evaluate fixed model queries at build time.
Ship only the offerings your application needs.
Save from the explorer/playground above via Save config.json,
generate with byokit.getConfigJson(), or write
byok.config.json:
{
"providerIds": [
"anthropic",
"google",
"openai",
"deepseek"
],
"local": false,
"filters": {
"includeTags": ["model:flash"]
}
}Evaluate your query at build time:
# Node / TS:
npx byokit build byok.config.json \
--out src/generated/models.js --compact
# Python:
byokit build byok.config.json \
--out app/generated_models.py --compactPair the generated model list with byokit/client:
// JavaScript / TypeScript:
import { getModels } from './generated/models.js';
import { call } from 'byokit/client';
const model = getModels().find(m =>
m.providerId === 'google');
if (!model) throw new Error('No Google model');
const key = process.env.GOOGLE_API_KEY;
if (!key) throw new Error('Set GOOGLE_API_KEY');
const result = await call(model, 'Hello', key);
byokit.getModels() queries a bundled snapshot offline. Precompiling a fixed
query removes the full catalog and query evaluation from your application’s runtime;
generated file size depends on the selected offerings.
06 / TRY A REAL CALL
Browse the complete list and model tags.
Verify an inference call with your own
key.
Test a model with your own key, then share a successful hosted-model report on GitHub. Earn a Model Verifier badge and help keep the catalog useful. Your key goes directly to the provider.
Verify a model · earn a badge ↗