krebsiti-ai

Official JavaScript client for the Krebsiti Developer API.

Install

npm install krebsiti-ai

Environment variables

export KREBSITI_API_KEY="your_api_key"
# optional:
export KREBSITI_BASE_URL="http://127.0.0.1:5001"
export KREBSITI_COLLECTION_ID="krb-your-collection-id"

Quick start

import { KrebsitiClient } from "krebsiti-ai";

const client = new KrebsitiClient({
  apiKey: process.env.KREBSITI_API_KEY!,
  baseURL: process.env.KREBSITI_BASE_URL, // optional
});

const completion = await client.chat.completions.create({
  model: "krebsiti-fast",
  messages: [{ role: "user", content: "გამარჯობა" }],
});

console.log(completion?.choices?.[0]?.message?.content || "");

Basic examples

1) Collections (KB-only)

const completion = await client.chat.completions.create({
  model: "krebsiti-fast",
  web_search: false,
  kb_collection_ids: [process.env.KREBSITI_COLLECTION_ID!],
  messages: [{ role: "user", content: "Summarize key points from the collection." }],
});

2) Web search

const completion = await client.chat.completions.create({
  model: "krebsiti-fast",
  web_search: {
    enabled: true,
    allowed_domains: ["matsne.gov.ge", "parliament.ge"],
  },
  messages: [{ role: "user", content: "რა არის უახლესი ცვლილებები შრომის კოდექსში?" }],
});

3) File search with Krebsiti collections

const collectionId = process.env.KREBSITI_COLLECTION_ID!;

const completion = await client.chat.completions.create({
  model: "krebsiti-fast",
  kb_collection_ids: [collectionId],
  file_search: {
    collection_id: collectionId,
    max_num_results: 5,
  },
  messages: [{ role: "user", content: "Find the section about deadlines and explain it." }],
});

4) Streaming

const stream = await client.chat.completions.create({
  model: "krebsiti-fast",
  stream: true,
  messages: [{ role: "user", content: "Give me a short summary." }],
});

for await (const chunk of stream as AsyncGenerator<any>) {
  const token = chunk?.choices?.[0]?.delta?.content;
  if (token) process.stdout.write(token);
}

5) Instructions (system behavior per request)

const completion = await client.chat.completions.create({
  model: "krebsiti-fast",
  instructions: [
    "You are Krebsiti.",
    "Always answer in Georgian (ka-GE).",
    "Keep answers short and practical.",
  ].join("\n"),
  messages: [{ role: "user", content: "Explain GDPR in 3 short bullets." }],
});

6) Streaming + web search + collections (with abort support)

const controller = new AbortController();

const stream = await client.chat.completions.create(
  {
    model: "krebsiti-fast",
    stream: true,
    kb_collection_ids: [process.env.KREBSITI_COLLECTION_ID!],
    web_search: { enabled: true },
    file_search: { collection_id: process.env.KREBSITI_COLLECTION_ID!, max_num_results: 5 },
    messages: [{ role: "user", content: "KB-first answer, then web sources if needed." }],
  },
  { signal: controller.signal },
);

Run local example files

node ./examples/01-collections.mjs
node ./examples/02-web-search.mjs
node ./examples/03-file-search-with-krebsiti-collections.mjs
node ./examples/04-stream.mjs
node ./examples/05-stream-with-web-and-collections.mjs
node ./examples/06-instructions.mjs

API compatibility

krebsiti-ai targets POST /api/v1/chat/completions and supports:

• model
• messages
• stream
• instructions
• temperature
• max_tokens
• web_search
• file_search
• kb_collection_ids

Errors

API failures throw KrebsitiApiError with:

• status
• code
• body