System onlineThe OS for AI work

Where AI systems
are built.

Zelaxy is the visual operating system for AI work — compose agents, workflows, automations, reasoning and knowledge on one canvas, then deploy to production.

LLM-agnostic/MCP-native/Local + Cloud/API-first/Open source
Canvas · runtime
Running· 1.28s
WebhookGmailMCP ToolsServersKnowledgeRAGMemoryVectorAgentOpus 4.8RouterLLM JudgeResponseAPISlackSend
Execution logrun #4821
01triggerwebhook/gmail · payload received
02agentagent · claude-opus-4-8 · planning
03toolmcp/github · list_issues → 12
04knowledgeknowledge · vector search → 6 hits
05memorymemory · upsert key=session
06routerrouter · llm-judge → escalate
07responserun #4821 · ok · 1.28s · $0.004
08
fig.00 — agent runtime graph8 nodes · 7 edges
§01Manifesto

The problem

AI got smart.
The tooling didn’t.

Models can reason, call tools, and write code. Yet most teams still wire them together with copy-pasted prompts, throwaway scripts, and dashboards no one trusts.

Intelligence is no longer the bottleneck. Building with it is.

Why existing tools fail

AChatbots

A prompt window is not a system. It forgets, it can’t act, and nothing is reproducible.

BGlue code

Scripts and SDKs rot. Every integration is bespoke, brittle, and invisible to your team.

CRigid automation

Linear pipelines can’t reason. The moment logic branches, they break.

The shift

AI systems are graphs of decisions, memory and tools. They should be built like systems — seen, versioned, and run — not hidden inside a chat box.

Zelaxy — premise

One visual canvas where humans and agents build, connect, reason, and deploy production AI — local or cloud, with every run observable.

BuildConnectReasonDeployScale
§02The Canvas

One canvas.
Every primitive.

Drag agents, models, memory, knowledge, MCP servers, logic and deploys onto an infinite canvas. Wire them together. Watch signals flow in real time.

workflow.zelaxy · main
x:1180 y:50015 nodes
+
ScheduleCronWebhookStripeVariablesStateKnowledgeRAGClassifierHaiku 4.5ResearcherSonnet 4.6MCP ToolsGitHubRouterLLM JudgeLoopforEachKnowledgepgvectorAgentGPT-5.4PostgresInsertResponseAPILogsTracesSlackSend
Node typesTriggersAgentsModelsRoutingLoopsKnowledgeMemoryMCPVectorsDataObserveDeploy+ 270 blocks
§03Lifecycle

From idea to
production, on
one surface.

No context-switching between a prompt playground, a code editor, and a scheduler. The whole loop lives on the canvas.

  1. 01

    Idea

    Start from a goal, a template, or a blank canvas.

  2. 02

    Drag

    Place agents, models, logic and integrations as blocks.

  3. 03

    Connect

    Wire data flow — branch, loop, and run paths in parallel.

  4. 04

    Knowledge

    Attach documents, vector stores and persistent memory.

  5. 05

    Reason

    Agents plan, call tools, and decide with LLM-as-judge.

  6. 06

    Test

    Run live and inspect every block, token by token.

  7. 07

    Deploy

    Ship as an API, webhook, schedule, or chat endpoint.

  8. 08

    Observe

    Trace every run with execution logs and metrics.

  9. 09

    Improve

    Version, branch, and iterate together with your team.

§04Capabilities

A complete
operating system.

Not a feature list — a coherent system. Every primitive a production AI team needs, on one surface, speaking one runtime.

01

Agent Block

LLM agents with tool use and structured output.

02

Visual Canvas

Infinite React Flow surface, drag & drop.

03

Workflow Engine

DAG executor with loops and parallel branches.

04

AI Copilot

The Wand builds workflows from a prompt.

05

Knowledge & RAG

pgvector search over your documents.

06

MCP Native

Connect any MCP server as live tools.

07

Memory

Persistent storage across runs and sessions.

08

Routing & Logic

Router, Condition, Loop, Parallel, Switch.

09

ZelaxyArena

Streaming chat engine with tool calls & cost.

10

Human in the Loop

Pause for review or approval, then resume.

11

Guardrails

Schema, regex, PII and hallucination checks.

12

Tables & Files

Built-in data tables and file storage.

13

Triggers & Schedules

Webhooks, cron, chat and manual runs.

14

Deployments

Ship as API, chat, or webhook endpoint.

15

Observability

Logs with trace spans, tool calls and cost.

16

Collaboration

Real-time multiplayer editing via Socket.IO.

§05Ecosystem

Not logos. Infrastructure.

Every model and service connects to one runtime bus. Swap providers without rewiring. Run frontier models or your own local stack — Zelaxy stays agnostic.

OpenAI
Anthropic
Google
xAI
DeepSeek
Mistral
Groq
Cerebras
Zelaxy Runtimev1.0
Bedrock
Azure
OpenRouter
Together
Fireworks
Ollama
vLLM
LiteLLM
GmailSlackGitHubNotionJiraStripeHubSpotAirtableLinearSalesforceDiscordTelegramGoogle SheetsGoogle DriveConfluenceSupabasePostgresDatabricksSnowflakeFirecrawlGmailSlackGitHubNotionJiraStripeHubSpotAirtableLinearSalesforceDiscordTelegramGoogle SheetsGoogle DriveConfluenceSupabasePostgresDatabricksSnowflakeFirecrawl

250+ integrations · MCP servers · custom tools

§06Enterprise

Built like
infrastructure.

Roles, secrets, environments and audit trails aren’t add-ons — they’re the substrate. Run Zelaxy in our cloud, your VPC, or fully on-prem.

Access
RBAC · read / write / admin
Secrets
AES-256 encrypted at rest
Isolation
Per-workspace multi-tenant
Deploy
Cloud · Self-host · On-prem
Audit
Immutable execution history
Auth
SSO · SAML 2.0 · OIDC · OAuth
0-bit
Secret encryption
0+
Integrations
0
Permission roles
0%
Self-hostable
PermissionsSingle Sign-OnTeamsProjectsAudit TrailsEncrypted SecretsPrivate DeploysEnvironmentsVersion ControlExecution HistoryObservabilityAnalyticsRate Limits
§07Developers

Every canvas is
just code.

What you build visually serializes to clean JSON you can diff, review, and version. Drop to the SDK, CLI or API whenever you want — nothing is locked behind the UI.

terminalworkflow.zelaxy.json
// workflow.zelaxy.json
{
  "id": "support-triage",
  "trigger": { "type": "webhook" },
  "blocks": [
    { "id": "classify", "tool": "agent", "model": "claude-haiku-4-5" },
    { "id": "route", "tool": "router", "paths": 4 },
    { "id": "reply", "tool": "gmail.send" }
  ],
  "edges": ["classify→route→reply"]
}

TypeScript SDK

Type-safe blocks, tools and triggers.

Python SDK

Drive workflows from notebooks or services.

REST + Streaming

Run any workflow over HTTP with SSE output.

Webhooks

Trigger from Gmail, GitHub, Stripe and more.

Custom Nodes

Ship your own blocks with full UI control.

Git Sync

Workflows are JSON — review them in PRs.

CLI

Scaffold, deploy and inspect from the terminal.

Open Source

MIT licensed. Read it. Fork it. Extend it.

§08Open Platform

Extend everything.
Own everything.

Zelaxy is open source and built to be forked. Write custom nodes, publish templates, wire in MCP servers, and ship your own integrations — no walled garden.

Custom ToolsCustom BlocksMCP ServersTemplatesSkillsOpen APIs
GitHub stars
0
Blocks
0+
Tools
0
Model providers

From the template registry

Support Triage

@core

Research Crew

@community

Data Pipeline

@community
System readyOpen source · MIT

Build the next
intelligent system.

Free to start. Deploy anywhere. Join the builders composing production AI on the canvas — local or cloud.

No credit card/Self-host in minutes/LLM-agnostic
Zelaxy — Visual AI Agent Builder & Workflow Automation