One event backbone. Zero hardcoded providers.
Every integration — inbound or outbound — speaks one language: CloudEvents 1.0. That single contract is what lets the catalog scale without ever rewriting the core engine.
Ingest
Agents and connectors normalize any source into standard CloudEvents on a Kafka backbone — the same envelope, whatever the origin.
Route
The workflow engine matches each event by pattern, cron, or webhook and runs the graph it triggers.
Act
Nodes call connectors, run AI agents, branch on conditions, or pause for a human — 42 node types in one canvas.
Compose
Every output is a new CloudEvent, so any automation can trigger any other. No point-to-point glue, ever.
42 workflow node types
Enough primitives to express real operations — triggers, control flow, data transforms, integration calls, and human steps.
Visual for operators. SDK for developers.
Non-developers compose on a drag-and-drop visual canvas. Builders extend the platform with a typed TypeScript SDK and CLI.
Visual Builder
A visual canvas with zoom, minimap, and live execution history — inspect, govern, and extend any template's underlying workflow.
TypeScript SDK & CLI
zervanor init | Scaffold a new agent project with manifest and boilerplate. |
zervanor dev | Run your agent locally with hot reload. |
zervanor push | Deploy to the platform over pure HTTP — no local Docker required. |
zervanor status | Check your project's deployment status on the platform. |
Tell it what you want. Zervanor builds it.
Describe your operation in plain language. Zervanor assembles a real, validated workflow — grounded in your actual connectors and agents — and shows it to you step by step before anything runs.
- Grounded in your live catalog — it selects real connectors/agents, never hallucinates one.
- Contract-validated — every step is checked against the canonical workflow schema, with in-place repair.
- Human-in-the-loop — it asks at real forks and never applies or deploys without your explicit approval.
Agent + Connector + Workflow = a ready Automation
An Automation is more than a bare workflow. It's a working business operation, assembled from three layers and deployed in one click.
Built for operators, hardened for production.
The platform handles the parts that usually stall a rollout — connecting accounts, and keeping integrations healthy.
Single-tap OAuth onboarding
The operator registers each provider's OAuth app once. Tenants then connect their accounts in a single tap — OAuth 2.0 and OAuth 1.0a, platform-managed, no per-tenant app setup.
Connector test & monitoring
Every connector runs periodic health tests with Prometheus metrics, Grafana dashboards, and Alertmanager incidents — plus a token-free public status page. Integrations don't silently rot.
Multi-tenant by construction
Per-tenant Kubernetes namespaces on GKE, org-scoped RBAC, and queries filtered by organization — validated by a passed security & governance audit (47/47 CI gates).
The catalog becomes a workforce.
The contract-native catalog was built for workflows. It turns out to be the skill library for an autonomous agent workforce — the same packs that power the templates auto-project into tools an AI agent can call.
- Generic Agentic Runtime shell — one provider-agnostic agent: manifest loader → tool bootstrap → ReAct planning loop. New agents are manifests + tools, not bespoke code.
- Tool projection — a connector's certified actions are exposed to agents over a tool API. Certify once → available to both workflows and agents.
- Bring your own model — OpenAI / Gemini today via a provider/base-url seam; the platform never hard-codes one vendor.
- Channel adapters — one agent serves many surfaces (chat, webhook, schedule).
Live today: the runtime shell, two agents on it (ai-agent + research-agent) with end-to-end goal execution, org BYOK LLM, and a management console. The 40-agent, 4-layer catalog is the sequenced build-out.