lxzl WORKFLOW ENGINE

Local-first workflow automation

lxzl

A visual automation console for building real data workflows: webhooks, schedules, AI steps, table transforms, notifications, integrations, and custom code, all on one inspectable canvas.

Trigger
Table Ops
Local AI
Notify

Why it exists

Automation should stay readable after it starts working.

lxzl is built for the moment when a workflow grows beyond a quick script. The graph shows the shape of the work, the config panel keeps details close to each node, and run history records exactly what happened when data moved through the system.

It also treats communication as part of the work, not a separate email black hole. Teams, vendors, clients, and partners should meet inside shared spaces where docs, boards, calendars, owners, and status are visible together.

It is approachable for non-programmers, but it does not hide the machinery from power users. Every node can be inspected, extended with a sandboxed script override, and connected to local or remote services through explicit credentials.

01

Visual DAG engine

Build with nodes and edges, run parallel branches, merge results, and watch status update on the canvas as the workflow executes.

02

Data tools built in

Load tables, parse CSV/JSON, filter, sort, group, join, pivot, deduplicate, stitch entities, and persist state between runs.

03

Local and hosted AI

Use Ollama by default, or connect Anthropic and OpenAI-compatible providers when a workflow needs hosted models.

04

Operational runtime

Schedule cron jobs, receive webhooks, queue work with retry, replay past runs, and monitor failures without adding Redis or a broker.

Architecture

One Bun process, one SQLite database, optional Ollama.

The current deployment model is deliberately compact: SvelteKit on Bun, Drizzle over bun:sqlite in WAL mode, a SQLite-backed job queue, Bun cron for schedules, and Bun Workers for isolated execution. Ollama can be managed by the app or run as a Docker sidecar.

That small shape keeps the system easy to run locally while still leaving room for cloud mode: authenticated workspaces, verified-email gates, credential vault encryption, and per-tenant database paths as the next scaling step.

61registered node types
200+starter templates
3AI provider paths
0Redis or broker dependencies

What it is good at

Workflows that need both clarity and escape hatches.

  • Webhook intake: receive events, validate payloads, branch by conditions, and return structured HTTP responses.
  • Data cleanup: normalize rows, merge sources, detect changes, and write summaries to internal tables.
  • AI enrichment: summarize tickets, classify leads, extract entities, and route results through deterministic nodes.
  • Ops automation: schedule recurring checks, send notifications, retry queued jobs, and inspect run history.
  • Integration work: generate API definitions from docs, attach encrypted credentials, and call actions from workflows.

Built With

Bun 1.3.14Runtime, package manager, SQLite, cron, Workers
Svelte 5Runes-based UI without a virtual DOM
SvelteKitRoutes, SSR, API handlers, app shell
SvelteFlowCanvas nodes, handles, edges, pan, zoom
SQLite + DrizzleWAL persistence and typed queries
OllamaManaged local model inference
ELK.jsWorker-backed automatic graph layout
MCPTool servers callable from workflows