AI Agents & MCPs & AI Workflow Automation • (~400 MCP servers for AI agents) • AI Automation / AI Agent with MCPs • AI Workflows & AI Agents • MCPs for AI Agents
TypeScript ai-agentai-agent-toolsai-agentsai-agents-frameworkmcpmcp-servermcp-toolsmcpsn8n-alternativeno-code-automationworkflowworkflow-automationworkflows
import { AIProviderName, isNil, tryCatch } from '@activepieces/core-utils'
import { AIProviderModelMetadata } from '@activepieces/shared'
import { z } from 'zod'
import { apLogger } from './ap-logger'
import { safeHttp } from './safe-http'
const logger = apLogger.create()
export const modelCatalog = {
async load(): Promise<ModelCatalogReader> {
const catalog = await loadCatalog()
return {
lookup({ provider, modelId }: { provider: AIProviderName, modelId: string }): AIProviderModelMetadata | undefined {
const source = CATALOG_SOURCE_PROVIDER[provider] ?? provider
const models = catalog?.providers[source]
if (isNil(models)) {
return undefined
}
return models[normalizeModelId({ provider: source, modelId })]
},
}
},
}
async function loadCatalog(): Promise<PublishedCatalog | undefined> {
if (!isNil(cached) && Date.now() - cached.fetchedAt < CATALOG_TTL_MS) {
return cached.value
}
if (!isNil(lastFailureAt) && Date.now() - lastFailureAt < FAILURE_BACKOFF_MS) {
return cached?.value
}
const pending = inFlight ?? startFetch()
const { data, error } = await tryCatch(() => pending)
if (!isNil(error)) {
return cached?.value
}
return data ?? undefined
}
function startFetch(): Promise<PublishedCatalog> {
inFlight = fetchCatalog()
.then((value) => {
cached = { value, fetchedAt: Date.now() }
lastFailureAt = undefined
return value
})
.catch((error) => {
lastFailureAt = Date.now()
logger.warn({ error, catalog: { url: catalogUrl() } }, 'Failed to load the AI model catalog; models will be returned without metadata')
throw error
})
.finally(() => {
inFlight = undefined
})
return inFlight
}
async function fetchCatalog(): Promise<PublishedCatalog> {
const response = await safeHttp.retryingAxios.get(catalogUrl(), {
timeout: REQUEST_TIMEOUT_MS,
})
return PublishedCatalog.parse(response.data)
}
function catalogUrl(): string {
return process.env['AP_MODEL_CATALOG_URL'] ?? DEFAULT_CATALOG_URL
}
function normalizeModelId({ provider, modelId }: { provider: AIProviderName, modelId: string }): string {
if (provider !== AIProviderName.BEDROCK) {
return modelId
}
return modelId.replace(BEDROCK_INFERENCE_PROFILE_PREFIX, '')
}
let cached: { value: PublishedCatalog, fetchedAt: number } | undefined
let inFlight: Promise<PublishedCatalog> | undefined
let lastFailureAt: number | undefined
const DEFAULT_CATALOG_URL = 'https://cdn.activepieces.com/ai/model-catalog.json'
const CATALOG_TTL_MS = 24 * 60 * 60 * 1000
const FAILURE_BACKOFF_MS = 5 * 60 * 1000
const REQUEST_TIMEOUT_MS = 10_000
const BEDROCK_INFERENCE_PROFILE_PREFIX = /^(us|eu|apac|global)\./
const CATALOG_SOURCE_PROVIDER: Partial<Record<AIProviderName, AIProviderName>> = {
[AIProviderName.ACTIVEPIECES]: AIProviderName.OPENROUTER,
}
const PublishedCatalog = z.object({
providers: z.partialRecord(z.enum(AIProviderName), z.record(z.string(), AIProviderModelMetadata)),
})
type PublishedCatalog = z.infer<typeof PublishedCatalog>
type ModelCatalogReader = {
lookup(params: { provider: AIProviderName, modelId: string }): AIProviderModelMetadata | undefined
}