Introduction

ExuluApp is the central class of an IMP backend. You instantiate it once, call the async create() method with your configuration and components, and then use the resulting instance to start the Express HTTP server and the BullMQ worker process. Everything the platform serves — agents, knowledge contexts, tools, evals, the MCP endpoint — is registered through this one object.

When to use it

Every IMP backend has exactly one ExuluApp. In the recommended three-file structure, src/exulu.ts creates the instance and exports the create() promise, while src/server.ts and src/worker.ts both await it before starting their process. You never construct a second instance; after create() resolves, the instance is also stored in an internal singleton so platform internals can reach it.

Minimal example

import { ExuluApp } from "@exulu/backend";

const app = new ExuluApp();

await app.create({
  config: {
    workers: { enabled: true },
    MCP: { enabled: true },
    telemetry: { enabled: false },
  },
  contexts: {},
  tools: [],
});

const server = await app.express.init();
server.listen(9001, () => {
  console.log("IMP backend listening on :9001");
});

create() returns a promise that resolves to the initialized ExuluApp. express.init() registers the GraphQL and REST routes, sets up authentication middleware, and returns the Express application you listen on.

What create() does

When you call app.create(), IMP:

Merges built-in contexts — the platform’s internal contexts (currently transcriptions) are merged after your custom contexts, so built-ins always win on key collision. A warning is logged if you define a context named transcriptions.
Registers 19 default providers — Anthropic (Claude Sonnet 4, Claude Sonnet 4.5, Claude Opus 4), OpenAI (GPT-5, GPT-5 Mini, GPT-5 Pro, GPT-5 Codex, GPT-5 Nano, GPT-4.1, GPT-4.1 Mini, GPT-4o, GPT-4o Mini), Google Vertex (Gemini 2.5 Flash, Gemini 2.5 Pro, Gemini 3 Pro, Llama Scout 4), and Cerebras (GPT-OSS 120B, Llama 3 8B, Llama 3.3 70B). Anything you pass in providers is appended alongside the defaults.
Loads built-in tools — todo, question, Perplexity, and email tools, plus the image-generation widget tool when LiteLLM is enabled and S3 file uploads are configured.
Registers default evals — when Redis is configured (REDIS_HOST + REDIS_PORT), the built-in LLM-as-judge eval is merged with any evals you pass. Without Redis, no evals are registered at all — including yours.
Validates IDs — every context, tool, and provider ID must start with a letter or underscore, use only letters, digits, and underscores, and be at most 80 characters. Treat 5 characters as the practical minimum. create() throws immediately on a violation.
Checks the queue license — if queues are registered without a valid EXULU_ENTERPRISE_LICENSE, create() throws. When Redis is reachable, the eval_runs queue is registered automatically.
Probes system dependencies — checks for the pandoc, soffice, and pdftoppm binaries and the globally installed docx npm package. By default create() throws when one is missing; set requireSystemDependencies: false to downgrade to a warning.

What the instance manages

Component	Passed via	Read back with
Contexts	`contexts` (record of `ExuluContext`)	`app.context(id)`, `app.contexts`
Providers	`providers` (array of `ExuluProvider`)	`app.provider(id)`, `app.providers`
Agents	`agents` (code-defined) + database	`await app.agent(id)`, `await app.agents()`
Tools	`tools` (array of `ExuluTool`)	`app.tool(id)`, `app.tools()`
Evals	`evals` (array of `ExuluEval`)	Registered for eval runs

Contexts are covered in depth in ExuluContext. Tools are covered in ExuluTool and providers in ExuluProvider.

Server and worker processes

The same ExuluApp instance powers two separate processes:

Server — await app.express.init() builds the Express application with all GraphQL and REST routes. When EXULU_USE_LITELLM=true it also starts the LiteLLM proxy supervisor, and when config.MCP.enabled is true it mounts the MCP endpoints under /mcp/:agent.
Worker — await app.bullmq.workers.create() connects to Redis and starts BullMQ workers for embeddings, processors, context sources, evals, and routines. This requires an Enterprise Edition license.

When to use it

Minimal example

What create() does

What the instance manages

Server and worker processes

Next steps

Configuration

API reference

​When to use it

​Minimal example

​What create() does

​What the instance manages

​Server and worker processes

​Next steps

Configuration

API reference

When to use it

Minimal example

What create() does

What the instance manages

Server and worker processes

Next steps