Adapter Integration

CtxIQ provides standardized connectors for every major LLM provider. Switch between OpenAI, Anthropic, and Llama without touching your session or context logic.

---

Available adapters

| Adapter ID | Provider | Context | Notes | | -------------------- | ----------- | ------- | -------------------------------------------------------------- | | openai-v4-turbo | OpenAI | 128k | Default. GPT-4 Turbo, recommended for most use cases. | | openai-v4o | OpenAI | 128k | GPT-4o multimodal — supports vision inputs. | | anthropic-claude-3 | Anthropic | 200k | Claude 3 family (Haiku, Sonnet, Opus). Largest context window. | | llama-3-70b | Meta / Groq | 8k | Ultra-low latency via Groq inference. | | llama-3-8b | Meta / Groq | 8k | Fastest, lowest cost. Good for high-volume classification. | | custom | Any | — | Bring your own via IAdapter interface. |

---

Usage

// Set adapter at initialization
const orchestrator = new CtxIQ({
  apiKey: process.env.CTXIQ_KEY,
  adapter: "anthropic-claude-3",
});

// Hot-swap adapter on an existing orchestrator
// (active sessions continue unaffected)
await orchestrator.setAdapter("llama-3-70b");

// Context logic is completely unchanged
const response = await session.ask("Same question, different model.");

---

Per-session adapter override

You can assign a different adapter to a specific session without affecting others:

const cheapSession = await orchestrator.createSession({
  id: "background-task",
  adapter: "llama-3-8b", // fast + cheap for low-stakes tasks
});

const qualitySession = await orchestrator.createSession({
  id: "user-facing",
  adapter: "anthropic-claude-3", // best quality for user interactions
});

---

Custom adapters

Implement the IAdapter interface to connect any inference endpoint — self-hosted models, fine-tunes, or edge deployments.

import { IAdapter, Message, CompletionResult } from "@ctxiq/sdk";

class MyAdapter implements IAdapter {
  readonly id = "my-custom-model";
  readonly tokenizer = "cl100k_base"; // used for accurate token counting

  async complete(messages: Message[]): Promise<CompletionResult> {
    const res = await fetch("https://my-model.example.com/v1/chat", {
      method: "POST",
      body: JSON.stringify({ messages }),
    });
    const data = await res.json();
    return {
      content: data.choices[0].message.content,
      tokens: data.usage.total_tokens,
    };
  }
}

// Register the custom adapter
const orchestrator = new CtxIQ({
  adapter: new MyAdapter(),
});

IAdapter interface

| Method | Required | Description | | ----------------------- | -------- | ------------------------------------------------------------------------------ | | complete(messages) | ✓ | Send messages and return content + token count. | | stream(messages, cb) | — | Optional streaming support. | | embed(text) | — | Optional — used for semantic pruning. Falls back to CtxIQ's embedding service. | | countTokens(messages) | — | Optional — override token counting. |

---