import { createRequire } from "node:module"; const require = createRequire(import.meta.url); const packageJson = require("../package.json"); import type { ExtensionAPI, ProviderModelConfig, } from "@earendil-works/pi-coding-agent"; import { OptionHelpers, Some } from "fp-sdk"; interface PPQPricing { input_per_1M_tokens: number; output_per_1M_tokens: number; } interface PPQArchitecture { modality: string; input_modalities: string[]; output_modalities: string[]; } interface PPQModel { id: string; name: string; context_length: number; pricing: PPQPricing; supported_parameters?: string[]; architecture?: PPQArchitecture; } interface PPQApiResponse { data: PPQModel[]; } const ppqApiBaseUrl = "https://api.ppq.ai"; function isMetaModel(modelId: string): boolean { const lowered = modelId.toLowerCase(); // e.g. AutoClaw and Auto return ( lowered.startsWith("auto") || // there's a bunch of free models in PPQ.ai website, maybe they'll get exposed by the API at some point? lowered.startsWith("free") ); } async function fetchPpqModels(): Promise { try { console.log("Fetching models from PPQ.ai..."); const response = await fetch(`${ppqApiBaseUrl}/v1/models`); if (!response.ok) { console.error( `Failed to fetch PPQ.ai models due to HTTP error; status: ${response.status}` ); return []; } const data = (await response.json()) as PPQApiResponse; console.log(`Fetched ${data.data.length} models from PPQ.ai`); return data.data; } catch (error) { console.error("Failed to fetch PPQ.ai models:\n", error); return []; } } async function filterPpqModels( apiModels: PPQModel[] ): Promise { try { const models: ProviderModelConfig[] = []; for (const model of apiModels) { const maybeSupportedParameters = OptionHelpers.ofObj( model.supported_parameters ); const supportedParameters = maybeSupportedParameters instanceof Some ? maybeSupportedParameters.value : []; const architecture = OptionHelpers.ofObj(model.architecture); // pi requires models to have tool support if ( !isMetaModel(model.id) && !supportedParameters.includes("tools") ) { continue; } let inputModalities: ("text" | "image")[] = ["text"]; if (architecture instanceof Some) { inputModalities = architecture.value.input_modalities.filter( (modality) => modality === "text" || modality === "image" ); } models.push({ id: model.id, name: model.name, api: "openai-completions", reasoning: supportedParameters.includes("reasoning"), input: inputModalities, cost: { input: model.pricing.input_per_1M_tokens, output: model.pricing.output_per_1M_tokens, cacheRead: 0, cacheWrite: 0, }, contextWindow: model.context_length, } as ProviderModelConfig); } const defaultModelId = "autoclaw"; const secondDefaultModelId = "auto"; models.sort((a, b) => { const position = (id: string) => { switch (id) { case defaultModelId: return 0; case secondDefaultModelId: return 1; default: return 2; } }; const diff = position(a.id) - position(b.id); if (diff !== 0) { return diff; } // alphabetically sort models that come after the defaults return a.id.localeCompare(b.id); }); console.log(`Found ${models.length} compatible models from PPQ.ai`); return models; } catch (error) { console.error("Failed to filter PPQ.ai models:\n", error); return []; } } export default async function (pi: ExtensionAPI) { console.log(`${packageJson.name} v${packageJson.version} initializing...`); const apiModels = await fetchPpqModels(); const models = await filterPpqModels(apiModels); if (models.length > 0) { pi.registerProvider("ppq", { baseUrl: ppqApiBaseUrl, api: "openai-completions", apiKey: "PPQ_API_KEY", models: models, }); console.log( `${packageJson.name} ready: Successfully loaded ${models.length} models from PPQ.ai` ); } else { console.error( `ERROR: no models from PPQ.ai could be fetched/configured` ); } return; }