Knott is a self-hosted workflow orchestration platform: AI-powered decisions, human-in-the-loop approval, a visual designer, real-time monitoring, and a full audit trail. Every AI decision runs on Cordon, our private inference engine, so models execute on your own hardware, all data stays inside your perimeter, and only a single port is ever exposed to the outside world.
Private inference · Human approval · Immutable audit · Single port
accountable
Sovereign by default
AI decisions run on Cordon on your own hardware. No data leaves your network unless you explicitly choose a cloud provider.
Human in the loop
Automatic escalation to human reviewers when AI confidence is low, with a required justification on every decision.
Immutable audit trail
Every AI decision, human approval, and node transition is recorded with timestamps and justification.
Runtime configurable
Switch providers, set models, and test connectivity from the Settings page. No restart, no editing files on the server.
Real-world triggers
Start workflows from inbound webhooks, not just manual runs. Any external system can POST JSON to kick off a run.
Lightweight footprint
Runs on a single self-hosted node with no external database and no heavyweight dependencies, so it is easy to deploy and easy to audit.
one exposed port
In production the Execution Engine serves the interface and reverse-proxies the registry, task, integration, and decision APIs. Clients see one port. Every other service stays bound to the internal network, and storage is local and self-contained, with no external database required. Scroll, and the topology powers on.
building blocks
Workflows are assembled in the visual designer from eight node types across four families. Each node is observable in the live run monitor, and each transition is written to the audit trail.
on Cordon
Knott routes every AI decision through Cordon, our private inference engine, so models execute on your own hardware, inside your perimeter, with no data egress. The same sovereign runtime that powers the rest of the Foundry powers Knott’s decision nodes.
Everything is configured at runtime: pick a provider, set the model, test the connection, and save. The choice persists across restarts with no editing files on the server, and when no engine is configured a deterministic rule-based simulation keeps testing and demos moving.
- →Private inference on your own hardware by default
- →Optional hosted provider where a cloud dependency is explicitly accepted
- →Deterministic simulation mode for testing without infrastructure
justified
A low-confidence AI decision does not slip through. It escalates to a human, waits for a justification, and the whole path, from trigger to outcome, lands in an immutable trail your auditors can read line by line.
Sovereign workflow orchestration with private inference on Cordon, human approval, and a complete audit trail, running entirely on your infrastructure.