Two engines. One design surface.
Design AI workflows and autonomous agents in one visual builder — then deploy the same artifact to cloud, on-prem, or air-gapped, with no modification and the audit trail intact.
- components
- 100+ components
- model integrations
- 22 model integrations
- credential providers
- 18 credential providers
- deploy target
- Any deploy targetcloud · on-prem · air-gapped
One platform, from canvas to production.
Two engines share the same components, credentials, and audit trail — so a workflow stays accountable from the moment you design it to the moment it runs in production.
Build visually
Compose workflows and agents from 100+ components in one drag-and-drop builder — a visual builder with code extensibility, so you can drop to code wherever you need finer control.
Run anywhere
Every workflow is a portable JSON artifact. Ship the same container image to cloud, on-prem, or air-gapped environments — with env-var differences only, zero modification.
Approve, audit, prove
Mark any step as critical and an agent pauses for human sign-off (beta). Every decision, approval, and credential access lands in the audit trail.
Structured automation, designed visually.
The Workflow Engine handles structured, repeatable work — document processing, RAG pipelines, data extraction, and scheduling — built on a drag-and-drop canvas with 100+ components across 22 model integrations. It’s a visual builder with code extensibility: drop to code wherever you need finer control.
100+ components
Prompts, models, retrievers, tools, logic, and I/O — chained on a visual canvas.
22 model integrations
OpenAI, Anthropic, Google, Groq, and more — behind one encrypted credential layer.
RAG & document pipelines
Chunk, embed, retrieve, and extract structured data over your own sources.
Scheduling & monitoring
Cron-based automation, with Prometheus, Grafana, and Loki observability.
Autonomous intelligence, with a human in the loop.
The Agent Engine adds autonomous reasoning — multi-step planning, conditional routing, cycles, and multi-agent collaboration — powered by production-grade orchestration. It shares the same components and credentials as the Workflow Engine, so the two compose: a pipeline can escalate an edge case to an agent, and an agent can call a structured workflow as a sub-task.
Reasoning & routing
Multi-step planning with conditional branches and cycles — not just linear chains.
Multi-agent collaboration
Coordinate specialised agents on a shared task, each with its own tools.
Human-in-the-loop gates
BETAPause at any critical step for reviewer sign-off, with the decision captured in the audit trail.
Streaming visibility
Watch agent state stream live over SSE as it plans, routes, and acts.
Shipping in beta. The Agent Engine and HITL approval gates are functional and demoable today. Durable, cross-restart execution is on the roadmap — not in yet — so in-progress approvals don’t survive a restart. See how we govern agents
One artifact. Every environment.
Where you design and where you run are independent. A workflow built in one MachineCraft instance is a portable artifact that runs in another on completely different infrastructure — your client’s requirements are never a blocker.
- Portable artifacts
- Workflows export as JSON and run on completely different infrastructure, unchanged.
- Same image, any target
- One container image deploys to public cloud, private cloud, on-prem, or air-gapped — env-var differences only.
- No phone-home
- The runtime needs no connection back to the design environment. Once deployed, it runs on its own.
Built on a proven open-source foundation.
MachineCraft extends a mature open-source workflow core with enterprise infrastructure and a second engine. Here’s what runs underneath.
- BACKEND
- Python · FastAPI
- DATABASE
- PostgreSQL
- WORKFLOW ORCHESTRATION
- LangChain
- AGENT ORCHESTRATION
- LangGraphBETA
- DEPLOYMENT
- Docker · Kubernetes
- OBSERVABILITY
- Prometheus · Grafana · Loki
- CREDENTIALS
- Fernet · AES-128-CBC
- API
- REST · v1 & v2
Approve, audit, prove.
The engines are how you build. Trust is why regulated teams can put them in production — approval gates, a full audit trail, and the governance layer we’re building next.
Design once. Deploy anywhere. Prove everything.
Join the private beta and be among the first teams to run AI agents you can actually put in production.