157 lines
5.4 KiB
TypeScript
157 lines
5.4 KiB
TypeScript
/**
|
|
* LLM Synthesizer — resolves the best available LLM for wiki page synthesis.
|
|
*
|
|
* Priority chain:
|
|
* 1. Anthropic Haiku (cheapest, fastest, best for synthesis)
|
|
* 2. Ollama (free, local)
|
|
* 3. Echo (no LLM, returns structured stub)
|
|
*
|
|
* All synthesizers implement LLMSynthesizeFn: (prompt: string) => Promise<string>
|
|
*/
|
|
|
|
import type { LLMSynthesizeFn } from './types.js';
|
|
|
|
export interface SynthesizerConfig {
|
|
/** Anthropic API key. Checked from env if not provided. */
|
|
anthropicApiKey?: string;
|
|
/** Ollama base URL. Checked from env if not provided. */
|
|
ollamaUrl?: string;
|
|
/** Ollama model name (default: llama3.2) */
|
|
ollamaModel?: string;
|
|
/** Max tokens for synthesis output (default: 1500) */
|
|
maxTokens?: number;
|
|
}
|
|
|
|
// ── Anthropic (Haiku) ───────────────────────────────────────────
|
|
|
|
function createAnthropicSynthesizer(apiKey: string, maxTokens: number): LLMSynthesizeFn {
|
|
return async (prompt: string): Promise<string> => {
|
|
// Dynamic import — @anthropic-ai/sdk is optional
|
|
const { default: Anthropic } = await import('@anthropic-ai/sdk');
|
|
const client = new Anthropic({ apiKey });
|
|
|
|
const response = await client.messages.create({
|
|
model: 'claude-haiku-4-5-20251001',
|
|
max_tokens: maxTokens,
|
|
messages: [{ role: 'user', content: prompt }],
|
|
});
|
|
|
|
const textBlock = response.content.find(b => b.type === 'text');
|
|
return textBlock?.text ?? '';
|
|
};
|
|
}
|
|
|
|
// ── Ollama ──────────────────────────────────────────────────────
|
|
|
|
function createOllamaSynthesizer(baseUrl: string, model: string, maxTokens: number): LLMSynthesizeFn {
|
|
return async (prompt: string): Promise<string> => {
|
|
const url = `${baseUrl.replace(/\/$/, '')}/api/generate`;
|
|
|
|
const response = await fetch(url, {
|
|
method: 'POST',
|
|
headers: { 'Content-Type': 'application/json' },
|
|
body: JSON.stringify({
|
|
model,
|
|
prompt,
|
|
stream: false,
|
|
options: { num_predict: maxTokens },
|
|
}),
|
|
signal: AbortSignal.timeout(60_000),
|
|
});
|
|
|
|
if (!response.ok) {
|
|
throw new Error(`Ollama error: ${response.status} ${response.statusText}`);
|
|
}
|
|
|
|
const data = await response.json() as { response: string };
|
|
return data.response ?? '';
|
|
};
|
|
}
|
|
|
|
// ── Echo (fallback) ─────────────────────────────────────────────
|
|
|
|
function createEchoSynthesizer(): LLMSynthesizeFn {
|
|
return async (prompt: string): Promise<string> => {
|
|
const frameMatch = prompt.match(/\((\d+) total\)/);
|
|
const frameCount = frameMatch ? frameMatch[1] : '?';
|
|
const nameMatch = prompt.match(/about "([^"]+)"/) ?? prompt.match(/concept "([^"]+)"/);
|
|
const name = nameMatch ? nameMatch[1] : 'this topic';
|
|
|
|
// Extract actual frame content for a basic summary
|
|
const frameLines = (prompt.match(/\[Frame #\d+.*?\]: .+/g) || []);
|
|
const facts = frameLines.slice(0, 8).map(line => {
|
|
const m = line.match(/\[Frame (#\d+).*?\]: (.+)/);
|
|
return m ? `- ${m[2].slice(0, 200)} *(${m[1]})*` : null;
|
|
}).filter(Boolean);
|
|
|
|
return `## Summary\nCompiled from ${frameCount} source frames about ${name}.\n\n` +
|
|
(facts.length > 0 ? `## Key Facts\n${facts.join('\n')}\n\n` : '') +
|
|
`> *Compiled with echo synthesizer. Connect an LLM for richer synthesis.*\n` +
|
|
`> Set ANTHROPIC_API_KEY or WAGGLE_OLLAMA_URL in your environment.`;
|
|
};
|
|
}
|
|
|
|
// ── Resolver ────────────────────────────────────────────────────
|
|
|
|
export interface ResolvedSynthesizer {
|
|
synthesize: LLMSynthesizeFn;
|
|
provider: 'anthropic' | 'ollama' | 'echo';
|
|
model: string;
|
|
}
|
|
|
|
/**
|
|
* Resolve the best available LLM synthesizer.
|
|
* Checks env vars and config, returns the first working option.
|
|
*/
|
|
export async function resolveSynthesizer(config?: SynthesizerConfig): Promise<ResolvedSynthesizer> {
|
|
const maxTokens = config?.maxTokens ?? 1500;
|
|
|
|
// 1. Try Anthropic
|
|
const anthropicKey = config?.anthropicApiKey
|
|
?? process.env.ANTHROPIC_API_KEY
|
|
?? process.env.WAGGLE_ANTHROPIC_API_KEY;
|
|
|
|
if (anthropicKey) {
|
|
try {
|
|
// Verify the SDK is importable
|
|
await import('@anthropic-ai/sdk');
|
|
return {
|
|
synthesize: createAnthropicSynthesizer(anthropicKey, maxTokens),
|
|
provider: 'anthropic',
|
|
model: 'claude-haiku-4-5-20251001',
|
|
};
|
|
} catch {
|
|
// SDK not available — fall through
|
|
}
|
|
}
|
|
|
|
// 2. Try Ollama
|
|
const ollamaUrl = config?.ollamaUrl ?? process.env.WAGGLE_OLLAMA_URL;
|
|
const ollamaModel = config?.ollamaModel ?? process.env.WAGGLE_OLLAMA_MODEL ?? 'llama3.2';
|
|
|
|
if (ollamaUrl) {
|
|
try {
|
|
// Quick health check
|
|
const health = await fetch(`${ollamaUrl.replace(/\/$/, '')}/api/tags`, {
|
|
signal: AbortSignal.timeout(3_000),
|
|
});
|
|
if (health.ok) {
|
|
return {
|
|
synthesize: createOllamaSynthesizer(ollamaUrl, ollamaModel, maxTokens),
|
|
provider: 'ollama',
|
|
model: ollamaModel,
|
|
};
|
|
}
|
|
} catch {
|
|
// Ollama not reachable — fall through
|
|
}
|
|
}
|
|
|
|
// 3. Echo fallback
|
|
return {
|
|
synthesize: createEchoSynthesizer(),
|
|
provider: 'echo',
|
|
model: 'echo',
|
|
};
|
|
}
|