Laava LogoLaava
Open Source · MIT License

We build the infrastructure we use. Now it's yours.

An open-source TypeScript SDK for building production AI agent platforms. Runtime, knowledge management, channels, observability — everything you need to go from idea to Kubernetes-ready deployment.

MIT License · TypeScript · Production-hardened

What's inside

Everything you need to go from idea to production agent platform. No boilerplate, no glue code.

Agent Runtime

Config loading with Zod validation, LLM client factories for OpenAI and Anthropic, JWT auth middleware with user identity mapping.

ConfigLLM WiringJWT Auth

Knowledge Management

Drizzle ORM database client, vector store abstractions for Qdrant and pgvector, document extractors for PDF, Word, Excel, and more.

DrizzleQdrantpgvectorExtractors

4 Built-in Channels

SSE for real-time streaming, Email with M365 and Gmail thread resolution, Slack with Block Kit, Teams with Adaptive Cards.

SSEEmailSlackTeams

Observability

Prometheus-compatible metrics with HTTP middleware, Langfuse tracing, Sentry error tracking, structured logging with Pino.

PrometheusLangfuseSentryPino

Review Workflows

Confidence gates for human-in-the-loop approval, review repository with SLA tracking, notifier integrations, decision audit trails.

HITLConfidence GatesSLA Tracking

CLI Scaffolder

Generate complete agent platforms with one command. Kubernetes manifests, CI/CD workflows, Docker configs, Helm charts — all wired up.

laava new platformK8sCI/CD

From zero to agent in minutes

The SDK handles config, LLM wiring, auth, and channel setup. You focus on what your agent actually does.

agent.ts
import { createConfig, createLogger } from '@laava-ai/sdk';
import { createLLMClient } from '@laava-ai/sdk/llm';
import { createSSEChannel } from '@laava-ai/sdk/channels';
import { createVectorStore } from '@laava-ai/sdk/db';

// Load config with environment validation
const config = createConfig();
const logger = createLogger(config);

// Wire up your LLM provider
const llm = createLLMClient(config, {
  provider: 'anthropic',
  model: 'claude-sonnet-4-20250514',
});

// Set up knowledge retrieval
const vectorStore = createVectorStore(config, {
  backend: 'qdrant',
});

// Expose via SSE channel
const channel = createSSEChannel(config, {
  llm,
  vectorStore,
  auth: { required: true },
});

channel.start();

Security-hardened. Sovereignty-ready.

Built with the same security defaults we use for enterprise deployments. No afterthoughts.

PII Redaction Built In

Logs and Sentry events automatically scrub 8+ credential and sensitive data types before they leave your environment.

JWT Authentication Required

Every /chat and /review route requires valid JWT tokens. No open endpoints by default.

Webhook Signature Verification

Slack webhooks fail-closed without valid signatures. M365 notifications are deduplicated with clientState validation.

Full Log Redaction

API keys, tokens, passwords, connection strings — all automatically stripped from logs and error reports.

More about our sovereign AI approach

Built by the team behind enterprise AI agent deployments

This SDK powers production systems for organizations across the Netherlands. Same code, same patterns, same security defaults. Need help building on it?

View our work

Frequently Asked Questions

View all questions

Start building

Install the SDK, scaffold a platform, and have an agent running in minutes. Or talk to our team about production deployment.

MIT License · @laava-ai/sdk · npm install laava

Laava SDK - Open-Source TypeScript SDK for AI Agent Platforms | Laava