Quick Start Guide

What Layerum Is

Layerum is a hosted RAG platform for teams that need to ingest knowledge, control access by workspace, and serve reliable answers through API or SDK.

Hosted endpoint: https://api.layerum.com.
Public integration contract: /v1/*.
Authentication: workspace API keys (lyr_wk_...).

Workspace Model

Every team works in isolated workspaces. Permissions, models, entities, documents, and API keys are controlled at workspace scope.

Multiple workspaces per account based on tier.
Team members, roles, and operational boundaries stay inside workspace.
API keys can be scoped by entities and model configs.

Data Model

Knowledge is organized to support both governance and retrieval precision.

Entity Types define high-level categories.
Entities form the hierarchy and retrieval scope.
Documents are attached to entities and indexed into chunks/vectors.

Ingestion Lifecycle

Documents pass through a predictable ingestion flow: upload, parse, chunk, embed, index, and status tracking.

Small files: direct multipart upload through API.
Large files: presigned flow (upload/init → S3 POST → upload/complete).
Ingestion statuses: pending, processing, completed, failed.
SDK helper ingestion.waitUntilReady handles polling for completion.

Capacity Control

Capacity is account-level by plan and can be distributed across workspaces dynamically. Enforcement is server-side for UI, API, and SDK flows.

Counted usage: entities, documents bytes, chunks, linked S3 files, pending upload reservations.
Not counted: audit/access logs, API key metadata, model/provider metadata.
Presigned uploads are hardened: reservation at upload/init, size/object validation at upload/complete, stale session cleanup.

Security and Governance

Access is explicit, scoped, and revocable at any time.

API keys are created and managed only in Layerum platform UI.
Capability model: query:read, data:ingest, and config:manage.
Per-key controls: entity scope, model scope, requests-per-minute, daily token limit, expiration.
Rotate and revoke are instant controls without changing integration architecture.

Retrieval Quality

Query behavior is designed for controllable and explainable answers, not just raw text generation.

Entity-scoped retrieval keeps context relevant.
Top-K and parent-scope options tune retrieval breadth.
Citations help trace generated answers back to source chunks.
Debug query endpoint supports deeper troubleshooting when needed.

Developer Experience

You can integrate with either typed SDK or plain HTTP while keeping the same public contract.

SDK package: @layerum-team/rag-sdk.
Public API base: https://api.layerum.com/v1.
Same auth and limits model for SDK and direct API clients.

import { LayerumClient } from '@layerum-team/rag-sdk';

const client = new LayerumClient({
  apiKey: process.env.LAYERUM_API_KEY!,
  baseUrl: 'https://api.layerum.com'
});

const source = await client.documents.uploadDirect({
  entityId: '<entity_uuid>',
  file,
  fileName: 'knowledge.pdf',
  contentType: 'application/pdf'
});

await client.ingestion.waitUntilReady({
  entityId: '<entity_uuid>',
  sourceId: source.id
});

const answer = await client.query.run({
  entityId: '<entity_uuid>',
  modelConfigId: '<model_config_uuid>',
  query: 'Summarize key risks and actions'
});

Operations and Reliability

The platform focuses on stable production behavior and observable failures.

Ingestion errors are persisted and available via API and SDK.
Retry endpoints support deterministic recovery flows.
Vector indexing is tied to document lifecycle for cleanup consistency.
Capacity checks reduce risk of runaway storage growth.

Typical Use Cases

Customer support knowledge assistants for internal teams.
Private documentation search across departments.
Compliance and policy retrieval with citations.
Team-scoped copilots connected to selected knowledge domains.

How to Start

1. Create a workspace and configure model connections.
2. Create API key in workspace UI and grant capabilities/scopes.
3. Create entity structure and upload documents.
4. Wait for ingestion completion, then run queries via SDK or API.

Continue with full endpoint and method details in Reference. Reference.

Platform Concepts