Configuration

StudioOptions configures targets around five boundaries: quick prompts, stores, UI, model catalog, and evaluation suites.

Stores

import { createSqliteSessionStore, Studio } from '@anvia/studio'

const store = createSqliteSessionStore({ path: '.anvia/studio.db' })

const studio = new Studio([agent, pipeline], {
  stores: {
    sessions: store,
    traces: store,
    pipelineLogs: store,
    pipelineRuns: store,
  },
})

The built-in SQLite implementation uses dedicated anvia_studio_* tables. Each surface has a public store contract, so applications can split stores or provide another backend. Sessions, pipeline logs, or pipeline runs can be disabled where their option accepts false; traces accept a trace store when custom persistence is required.

Models

Register providers with stable IDs and a createCompletionModel factory. Optional static definitions and listModels populate the catalog. Per-agent policies define defaults and allowed provider:model references.

Model policy is enforced by the Studio runtime; a request outside an agent's allowed set is rejected. The factory still owns provider credentials and provider-specific construction.

UI

const studio = new Studio([agent], {
  ui: {
    path: '/playground',
    title: 'Support Studio',
    rootRoutes: true,
    redirectRoot: true,
    clientScript: `console.info('Studio client loaded')`,
  },
})

path, rootRoutes, and redirectRoot change route presentation. clientScript is JavaScript source served as the Studio client module, so only provide trusted code. protectShell is a public option but currently does not authenticate users or protect runtime APIs.

Serving

start() launches a Node HTTP server and returns immediately. If port is omitted, Studio checks RUNNER_PORT and then uses 4021. Use fetch(request) or the exposed Hono app when another server owns request dispatch.

See runtime boundary and UI/server options.