@khaveeai/providers-mock

Canned, randomized MockLLM/MockTTS classes for exercising chat/animation-trigger logic locally — no API keys, no network calls, no cost.

Install

npm install @khaveeai/providers-mock

What this is

• MockLLM implements LegacyLLMProvider's streamChat({messages}) => AsyncIterable<{type, delta}>. It matches the last message against a fixed set of keywords (hello, dance, sad, happy, angry, think, yes, no, ...) or falls back to one of 8 canned responses, then yields it character-by-character with a randomized 20–80ms delay to simulate typing.
• MockTTS implements LegacyTTSProvider's speak({text, voice?}) => Promise<void>. It does not play audio — it logs to console and awaits a fake duration (capped at 8s) computed from word count.

⚠️ This does not plug into the avatar/voice pipeline. LegacyLLMProvider/LegacyTTSProvider are an older interface pair, separate from RealtimeProvider — the interface KhaveeProvider/useRealtime actually consume to drive the avatar. KhaveeConfig does have optional llm/tts fields typed for these, so <KhaveeProvider config={{ llm: new MockLLM(), tts: new MockTTS() }}> compiles — but nothing reads config.llm/config.tts anywhere in @khaveeai/react. No chat status, no lip-sync, no avatar reaction.

What it's actually useful for

Standalone testing of chat/animation-trigger logic, independent of the avatar:

• Unit-test that streamChat yields the expected *trigger_animation: name* marker for a given keyword.
• Sanity-check your own trigger-parsing logic before wiring a real LLM.
• Preview what a "typing" delta stream looks like, with zero LLM cost.

Triggers like *trigger_animation: wave_small* are plain text in the canned responses — parse them yourself (e.g. text.match(/\*trigger_animation:\s*(\w+)\*/)); MockLLM/MockTTS don't act on them.

Usage

import { MockLLM, MockTTS } from "@khaveeai/providers-mock";

const llm = new MockLLM();
let response = "";
for await (const chunk of llm.streamChat({ messages: [{ role: "user", content: "hello there" }] })) {
  if (chunk.type === "text") response += chunk.delta;
}
console.log(response);
// => "Hello there! Nice to meet you! *trigger_animation: wave_small* 👋"

const tts = new MockTTS();
await tts.speak({ text: response, voice: "mock-voice" }); // logs + waits, plays nothing

Want the real avatar + lip-sync pipeline without API keys?

MockLLM/MockTTS aren't the path for that. Either write a minimal class implementing RealtimeProvider directly, or use @khaveeai/providers-generic-stt-tts's GenericPipelineProvider with your own stub VAD/STT/LLM/TTS adapters — either gives KhaveeProvider a real RealtimeProvider to drive the avatar end-to-end.

License

MIT