lxzl WORKFLOW ENGINE
BUSINESS AUTOMATION, MADE VISIBLE

Workflows you can see thinking.

lxzl makes every run visible: fields, branches, AI decisions, credentials, and business outcomes in one self-hostable workflow console.

customer.name deal.stage ai.intent next.action
50+ node surfaces
Local + cloud AI routing
Self-host Bun + SQLite

No credit card Self-host ready Local AI in dev, cloud AI in prod

THE BUSINESS LAYER

Expressive design. Operational clarity.

The visuals are there to make complex flow state understandable: what came in, what changed, which branch won, and which system got updated.

Capture

Bring every event onto one canvas

Forms, webhooks, tables, CRMs, calendars, and APIs enter the same visible workflow map.

Understand

Make data readable while it moves

Field references, live payloads, schema propagation, and node previews keep users oriented.

Decide

Let AI work where it earns its keep

Use local Ollama in dev, route to OpenAI or Anthropic in prod, and keep tool calls auditable.

Ship

Turn workflows into business outcomes

Update systems, notify teams, schedule runs, replay history, and expose workflows as MCP tools.

Qualify inbound leadsSync customer recordsGenerate internal reportsRoute approvalsEnrich tables with AITrigger ops alerts
CAPABILITIES

Everything a workflow needs.

A dense toolkit built for real pipelines — not a demo.

40+ node types

Triggers, conditions, loops, filters, HTTP, AI, email, notifications — everything a pipeline needs.

Click-to-configure

Pick fields from dropdowns, compose conditions visually. No syntax to memorize.

Local AI, zero keys

Built-in Ollama runs Gemma, Qwen, Llama locally. Works offline, no API bills.

AI Agents with tools

Multi-turn agent node calls MCP + HTTP tools until done. Structured JSON output, chat memory.

Live data preview

Watch every node light up as data flows. Inspect the exact payload at each step.

Script Override

Every node has a JS escape hatch. Sandboxed Worker, no restrictions on logic.

Cron + webhooks

OS-level cron with full expression support. HMAC-signed webhooks for any external source.

MCP plugins

Connect any Model Context Protocol server — tools become first-class nodes on the canvas.

Encrypted vault

AES-256-GCM credential store with per-role grants. Admins hold keys; editors just use them.

Run history + replay

Every execution captured. Scrub the timeline, replay past runs, filter by status or date.

HOW IT WORKS

From idea to production in three steps.

01

Drag & connect

Drop nodes, wire them up, configure each with dropdowns and pickers. No YAML, no boilerplate.

02

Run with live preview

Hit run, watch data flow through every node in real time. Inspect, fix, iterate in seconds.

03

Ship it

Schedule with cron, trigger via webhook, or call it as an MCP tool. Or batch-run from CSV.

WHAT PEOPLE BUILD

Four categories. Endless specifics.

Data pipelines

ETL between APIs, transform with visual filters, enrich with AI, load into tables or webhooks.

AI agents

Chat triggers, tool-calling LLMs, conversation memory. Structured output with JSON schemas.

CRM & marketing

Lead scoring, drip campaigns, form handlers, notification fan-out. Condition-driven routing.

Internal automation

Scheduled reports, approval workflows, sync between systems. Run it yourself — no SaaS fees.

UNDER THE HOOD
BUN · RUNTIMESVELTE 5 · UISVELTEFLOW · CANVASSQLITE · STORAGEOLLAMA · LOCAL AIMCP · PLUGINS
QUESTIONS

Things people ask.

Can I self-host?

Yes — lxzl is a single Bun process with SQLite. No Kubernetes, no Redis, no external services required. Docker + docker-compose files included.

Does it work offline?

The whole editor runs in the browser talking to your local Bun server. Ollama runs models on your machine. No outbound calls unless you configure an external AI provider.

How is this different from n8n or Zapier?

lxzl is single-binary self-hostable with built-in local AI, a first-class AI Agent node with MCP tools, and an MCP server mode exposing your workflows to Claude Desktop / Cursor. n8n lacks the MCP server + local-AI defaults; Zapier is SaaS-only.

Is my data secure?

Credentials are AES-256-GCM encrypted. SSRF guard on all outbound HTTP. Per-org tenant isolation. Script overrides run in Bun Workers with no access to fetch, process, or the filesystem.

Which AI models are supported?

Ollama (local, default — Gemma, Qwen, Llama, Mistral), Anthropic (Claude 4.x), and any OpenAI-compatible API (OpenAI, Groq, Together, vLLM). Tool-calling on all three.

Can I extend it?

Every node has a Script Override toggle. Flip it, write custom JS, keep the visual config. Also: register custom integrations from an OpenAPI spec, or add your own MCP server.

READY

Ready to automate?

Sign up free. Your first workflow runs in a minute.

Create your account →