@kitana-sdk/core

Router, provider detection, and failover for Kitana SDK — use AI subscriptions (Claude Max, ChatGPT Plus) programmatically, without API keys.

Install

npm install @kitana-sdk/core

Usage

import { createRouter, detect } from '@kitana-sdk/core'

const env = await detect()
// { providers: { claude: {...}, codex: { available, auth: {...} }, ollama: {...} }, ... }

const router = createRouter({
  chain: ['claude', 'codex', 'ollama', 'api-key'],
  models: { ollama: 'mistral:instruct', claude: 'sonnet' },
  apiKeys: { anthropic: process.env.ANTHROPIC_API_KEY }
})

const response = await router.complete({
  messages: [{ role: 'user', content: 'Hello' }],
  model: 'auto',
  systemPrompt: 'Answer concisely.'
})
// { content, model, provider, usage }

models configures a default per provider when the request model is auto or omitted. An explicit request model still takes precedence. Providers without a configured model retain their existing defaults.

Streaming

import { streamClaude } from '@kitana-sdk/core'

await streamClaude('say PONG', 'auto', text => {
  process.stdout.write(text) // called incrementally as tokens arrive
})

Provider fallback context

onProviderSwitch lets a caller (e.g. @kitana-sdk/bible) inject context when the router falls back to a different provider. The context is passed through each provider's real system-prompt channel — never mixed into the user message — so it isn't mistaken for a prompt-injection attempt.

const router = createRouter({
  chain: ['claude', 'ollama'],
  onProviderSwitch: async ({ from, to }) => {
    return 'facts carried over from the previous step...'
  }
})

Requirements

• Claude CLI (npm install -g @anthropic-ai/claude-code), authenticated via claude auth login
• Codex CLI (npm install -g @openai/codex), authenticated via codex login
• Optionally: Ollama running locally, or an Anthropic/OpenAI API key for fallback

See the main repo for full documentation.