The canvas doesn't disappear — it evolves
Stop wiring 50 nodes by hand. Describe intent in one sentence — supervisor, provisioning, security, observability, and FinOps agents spawn on demand, plan their own routing, and render the live execution DAG straight onto the canvas.
# This replaces an entire n8n workflow workflow = AgentWorkflow( intent=""" For every new tenant onboarded: - Provision isolated VPC + EKS cluster on AWS - Run CIS benchmark security scan - Deploy Prometheus + Grafana observability - Tag all resources with tenant cost center - Email provisioning report to tenant admin - Set SLO alerts and escalation policies """, tenant_id="acme-corp", canvas=True # renders live execution DAG )workflow.run()
1
sentence of intent
N
agents spawn on demand
50×
less wiring vs n8n
Live
execution DAG
The shift
Manual DAG vs Agentic Workflow
You stop being the wiring engineer and become the intent author. Agents take it from there.
Meet the crew
The agents that show up when you need them
Each agent has a sharp scope, its own tools, and a budget. The supervisor spawns only the ones your intent actually requires.
SupervisorAgent
Decomposes intent into a plan, routes work to specialist agents, and re-plans on failure.
ProvisioningAgent
Calls Terraform, Pulumi, or cloud APIs to create VPCs, clusters, databases, and load balancers.
SecurityAgent
Runs CIS / SOC 2 / HIPAA benchmarks, scans IaC for drift, and gates promotions on policy.
ObservabilityAgent
Wires Prometheus, Grafana, Loki, and OTel exporters; defines SLOs and on-call routing.
FinOpsAgent
Tracks cost per node, applies tags, flags overspend, and rewrites plans to hit budget.
Human-in-loopAgent
Pauses on configurable risk thresholds — production writes, IAM changes, paid resources.
Lifecycle
From sentence to live DAG in 4 phases
Why it works in production
Agentic, but accountable
Every agent action is observable, gated, and reversible. You get autonomy without the "why did the bot do that" investigation.
Self-routing plans
Supervisor decomposes goals into agent calls — no manual edges, no drag-and-drop migrations.
Self-healing on failure
Agents retry with alternate strategies and replay the new path on the canvas in real time.
Full execution audit
Every agent call, tool invocation, and decision is captured, hash-chained, and exportable to your SIEM.
Per-step approval gates
Configurable HITL checkpoints for production writes, IAM changes, and paid cloud resources.
Bring your own models
Route per-agent between Claude, GPT, Gemini, or self-hosted Llama / Mistral via VxAI gateway.
FinOps annotations
Cost delta is appended to every node — workflow spend is observable, not estimated after the fact.
Agents already know how to talk to
Stop wiring nodes. Start declaring intent.
Spin up your first agentic workflow in under a minute. Free tier includes the supervisor, 3 specialist agents, and a live canvas.