/** * Generic provider client factory. * * Given a ProviderEntry from the store, returns a ProviderClient that calls * the right SDK based on apiType. Three dispatch paths: * * - anthropic: @anthropic-ai/sdk (native Messages API) * - gemini: @google/generative-ai (native) * - openai-compat: openai SDK with a custom baseURL — covers OpenAI, * DeepSeek, Mistral, Groq, OpenRouter, xAI, Together, … * * The mock provider short-circuits everything when MOD8_MOCK=1. */ import Anthropic from '@anthropic-ai/sdk'; import OpenAI from 'openai'; import { GoogleGenerativeAI } from '@google/generative-ai'; import type { ProviderClient, ProviderResponse, ProviderCallOptions, StreamEvent, } from './types.js'; import { priceFor } from './pricing.js'; import { resolveConfigured, type ProviderEntry } from '../storage/providers.js'; import { templateById } from './registry.js'; import { mockProvider } from './mock.js'; import { resolveModel } from './modelResolution.js'; import { debugProviderCall, debugProviderError, debugProviderResponse, } from '../util/debug.js'; /** * Build a ProviderClient bound to a specific configured provider id. * Throws a friendly error if the provider isn't configured. */ export async function buildProviderClient(id: string): Promise { if (process.env.MOD8_MOCK === '1') return mockProvider(id); const entry = await resolveConfigured(id); if (!entry) { const tpl = templateById(id); const label = tpl?.name ?? id; throw new Error( `No ${label} key configured.\n` + ` · mod8 login (recommended — one paste connects all four built-in providers)\n` + ` · mod8 keys set ${id}` + (tpl ? '' : `\n · mod8 add-provider (for custom OpenAI-compatible endpoints)`) + '.' ); } return clientForEntry(id, entry); } function clientForEntry(id: string, entry: ProviderEntry): ProviderClient { switch (entry.apiType) { case 'anthropic': return anthropicClient(id, entry); case 'openai-compat': return openaiCompatClient(id, entry); case 'gemini': return geminiClient(id, entry); } } // ---------- Anthropic ---------- function anthropicClient(id: string, entry: ProviderEntry): ProviderClient { const client = new Anthropic({ apiKey: entry.apiKey }); return { id, defaultModel: entry.defaultModel, async call(prompt: string, opts: ProviderCallOptions = {}): Promise { const resolved = resolveModel(id, opts.model, entry.defaultModel); const model = resolved.model; debugProviderCall({ providerId: id, apiType: 'anthropic', model, modelSource: resolved.source, apiKey: entry.apiKey, promptPreview: prompt, }); const start = Date.now(); try { const res = await client.messages.create({ model, max_tokens: opts.maxTokens ?? 1024, messages: [{ role: 'user', content: prompt }], }); const latencyMs = Date.now() - start; const text = res.content .filter((b: { type: string }) => b.type === 'text') .map((b: unknown) => (b as { text: string }).text) .join(''); const inputTokens = res.usage.input_tokens; const outputTokens = res.usage.output_tokens; const actualModel = res.model ?? model; debugProviderResponse(id, actualModel, { input: inputTokens, output: outputTokens, latencyMs }); return { text, inputTokens, outputTokens, costUsd: priceFor(actualModel, inputTokens, outputTokens), latencyMs, model: actualModel, }; } catch (err) { debugProviderError(id, err); throw err; } }, async *stream(prompt: string, opts: ProviderCallOptions = {}): AsyncIterable { const resolved = resolveModel(id, opts.model, entry.defaultModel); const model = resolved.model; debugProviderCall({ providerId: id, apiType: 'anthropic', model, modelSource: resolved.source, apiKey: entry.apiKey, promptPreview: prompt, }); const start = Date.now(); try { const ms = client.messages.stream({ model, max_tokens: opts.maxTokens ?? 1024, messages: [{ role: 'user', content: prompt }], }); for await (const ev of ms) { if (ev.type === 'content_block_delta' && ev.delta.type === 'text_delta') { yield { type: 'text', delta: ev.delta.text }; } } const final = await ms.finalMessage(); const latencyMs = Date.now() - start; const inputTokens = final.usage.input_tokens; const outputTokens = final.usage.output_tokens; const actualModel = final.model ?? model; debugProviderResponse(id, actualModel, { input: inputTokens, output: outputTokens, latencyMs }); yield { type: 'done', usage: { inputTokens, outputTokens, latencyMs, model: actualModel, costUsd: priceFor(actualModel, inputTokens, outputTokens), }, }; } catch (err) { debugProviderError(id, err); throw err; } }, }; } // ---------- OpenAI-compatible (OpenAI, DeepSeek, Mistral, Groq, ...) ---------- function openaiCompatClient(id: string, entry: ProviderEntry): ProviderClient { const client = new OpenAI({ apiKey: entry.apiKey, baseURL: entry.baseUrl, }); return { id, defaultModel: entry.defaultModel, async call(prompt: string, opts: ProviderCallOptions = {}): Promise { const resolved = resolveModel(id, opts.model, entry.defaultModel); const model = resolved.model; debugProviderCall({ providerId: id, apiType: 'openai-compat', model, modelSource: resolved.source, baseUrl: entry.baseUrl, apiKey: entry.apiKey, promptPreview: prompt, }); const start = Date.now(); try { const res = await client.chat.completions.create({ model, messages: [{ role: 'user', content: prompt }], max_tokens: opts.maxTokens ?? 1024, }); const latencyMs = Date.now() - start; const text = res.choices[0]?.message?.content ?? ''; const inputTokens = res.usage?.prompt_tokens ?? 0; const outputTokens = res.usage?.completion_tokens ?? 0; const actualModel = res.model ?? model; debugProviderResponse(id, actualModel, { input: inputTokens, output: outputTokens, latencyMs }); return { text, inputTokens, outputTokens, costUsd: priceFor(actualModel, inputTokens, outputTokens), latencyMs, model: actualModel, }; } catch (err) { debugProviderError(id, err); throw err; } }, async *stream(prompt: string, opts: ProviderCallOptions = {}): AsyncIterable { const resolved = resolveModel(id, opts.model, entry.defaultModel); const model = resolved.model; debugProviderCall({ providerId: id, apiType: 'openai-compat', model, modelSource: resolved.source, baseUrl: entry.baseUrl, apiKey: entry.apiKey, promptPreview: prompt, }); const start = Date.now(); try { const stream = await client.chat.completions.create({ model, messages: [{ role: 'user', content: prompt }], max_tokens: opts.maxTokens ?? 1024, stream: true, stream_options: { include_usage: true }, }); let inputTokens = 0; let outputTokens = 0; let actualModel = model; for await (const chunk of stream) { const delta = chunk.choices[0]?.delta?.content; if (delta) yield { type: 'text', delta }; if (chunk.usage) { inputTokens = chunk.usage.prompt_tokens ?? 0; outputTokens = chunk.usage.completion_tokens ?? 0; } if (chunk.model) actualModel = chunk.model; } const latencyMs = Date.now() - start; debugProviderResponse(id, actualModel, { input: inputTokens, output: outputTokens, latencyMs }); yield { type: 'done', usage: { inputTokens, outputTokens, latencyMs, model: actualModel, costUsd: priceFor(actualModel, inputTokens, outputTokens), }, }; } catch (err) { debugProviderError(id, err); throw err; } }, }; } // ---------- Gemini ---------- function geminiClient(id: string, entry: ProviderEntry): ProviderClient { const genAI = new GoogleGenerativeAI(entry.apiKey); return { id, defaultModel: entry.defaultModel, async call(prompt: string, opts: ProviderCallOptions = {}): Promise { const resolved = resolveModel(id, opts.model, entry.defaultModel); const modelName = resolved.model; debugProviderCall({ providerId: id, apiType: 'gemini', model: modelName, modelSource: resolved.source, apiKey: entry.apiKey, promptPreview: prompt, }); const model = genAI.getGenerativeModel({ model: modelName }); const start = Date.now(); try { const result = await model.generateContent(prompt); const latencyMs = Date.now() - start; const text = result.response.text(); const usage = result.response.usageMetadata; const inputTokens = usage?.promptTokenCount ?? 0; const outputTokens = usage?.candidatesTokenCount ?? 0; debugProviderResponse(id, modelName, { input: inputTokens, output: outputTokens, latencyMs }); return { text, inputTokens, outputTokens, costUsd: priceFor(modelName, inputTokens, outputTokens), latencyMs, model: modelName, }; } catch (err) { debugProviderError(id, err); throw err; } }, async *stream(prompt: string, opts: ProviderCallOptions = {}): AsyncIterable { const resolved = resolveModel(id, opts.model, entry.defaultModel); const modelName = resolved.model; debugProviderCall({ providerId: id, apiType: 'gemini', model: modelName, modelSource: resolved.source, apiKey: entry.apiKey, promptPreview: prompt, }); const model = genAI.getGenerativeModel({ model: modelName }); const start = Date.now(); try { const result = await model.generateContentStream(prompt); for await (const chunk of result.stream) { const text = chunk.text(); if (text) yield { type: 'text', delta: text }; } const final = await result.response; const usage = final.usageMetadata; const inputTokens = usage?.promptTokenCount ?? 0; const outputTokens = usage?.candidatesTokenCount ?? 0; const latencyMs = Date.now() - start; debugProviderResponse(id, modelName, { input: inputTokens, output: outputTokens, latencyMs }); yield { type: 'done', usage: { inputTokens, outputTokens, latencyMs, model: modelName, costUsd: priceFor(modelName, inputTokens, outputTokens), }, }; } catch (err) { debugProviderError(id, err); throw err; } }, }; }