Regnant
Products/02 · KnottDeployment Ready

Knott.

Automation you can actually be accountable for.

Self-hosted, AI-powered workflow automation with human-in-the-loop decisions, a visual designer, real-time execution monitoring, and a full audit trail. The power of a modern orchestration platform, with every AI decision running locally and all data staying on your own infrastructure.

The premise

Automate the routine. Keep a human on the consequential. Record everything, with its justification.

Regnant · SYSTEM RECORDSystem02 / 07
justification attached↖ confidence gate ◇↖ the human ○
● step: recorded with justification◇ junction: confidence gate○ the human: pulled in, on the record
Plate II: one route, every step on the recordRegnant · Sovereign AI · Dar es Salaam

One run, drawn to scale

The shape of a decision.

Hover a node. The green edge is machine confidence; the orange one is a human being pulled in.

T1, Trigger

API/webhook entry point: workflows start from real-world events, not just manual runs.

The moment that matters

Feel the handoff.

Escalation isn't a feature bullet: it's a line. Drag the model's confidence across it and watch who becomes responsible.

91%
policy line · 75%

Machine lane

Confidence clears the line: the run proceeds and records its own justification.

Human lane

Confidence drops below the line: the run pauses here, with you, and an SLA clock starts.

The audit trail: every completion, either way, lands here

  • Nothing recorded yet. Complete a run: machine or human, and read what Knott writes down.

100% of consequential steps recorded with justification, including the machine's.

Node reference: 8 types

01

Trigger

API/webhook entry point: workflows start from real-world events, not just manual runs.

02

AI Decision

Confidence-scored, model-selectable decision step running on your chosen provider.

03

Human Task

Pauses for review or approval, with SLA tracking. Low AI confidence escalates here automatically.

04

Condition

Branches the run on expressions.

05

Tool Call

Real connectors: webhook, Slack, SendGrid, Twilio. Credentials come from environment variables, never the workflow definition.

06

Agent Call

Calls a registered external agent from the agent registry.

07

Parallel

Fan-out / fan-in across simultaneous branches.

08

End

Terminal outcome: Approve, Reject, Complete, or Escalate.

Surface area

Five services.
One port out.

In production, the Execution Engine is the single reverse proxy for everything else. Internal service ports stay bound to the host network. The outside world sees exactly one opening.

:443

Workflow Registry

Execution Engine← the one exposed

AI Decision Engine

Human Task Service

Agent Integration

Everything inside the wall. One gap, on your terms.

Choose your intelligence: switchable at runtime

Sovereignty

Ollama

Local models, fully offline, cost under control. No data egress. GPU recommended.

Quality

Anthropic Claude

Highest decision quality. Requires an API key and internet, your call, made explicitly.

Determinism

Simulation

Rule-based fallback, always available. Deterministic runs for testing and demos.

On the record

  • Every AI decision and human approval recorded with timestamp and justification
  • Automatic escalation to human reviewers when AI confidence is low
  • Connector credentials read from environment variables at runtime, never stored in a workflow definition
  • Runs behind your reverse proxy; TLS and authentication terminate at your layer
  • Runtime-configurable AI: switch providers and test connectivity with no restart
  • Full light/dark theming, switchable per device

Knott preview

Video placeholder