HestiaOS

What is HestiaOS?

Problem

Agentic systems — whether AI assistants, autonomous workflows, or multi-agent architectures — face fundamental challenges around decision transparency, accountability, and boundary control.

When an agent acts, several questions arise:

  • Who decided the action should proceed?
  • What policy governed the decision?
  • What evidence exists that the decision was valid?
  • Can the chain of events be reconstructed later?
  • What side effects did the action produce?

Most existing agent frameworks do not systematically answer these questions. Policy is often implicit, traces are incomplete, and there is no clear line between "proposed" and "executed".

Approach

HestiaOS explores a governance-first execution model.

Instead of treating policy as an afterthought or guardrail bolted onto an otherwise unrestricted agent, HestiaOS places a governance kernel at the boundary between intent and side effect.

This kernel is responsible for:

  • Registering incoming intents
  • Evaluating them against explicit policy
  • Producing DecisionTraces — replayable, causal records
  • Requiring an ExecutionCommit before any external side effect is permitted

Core Concepts

ConceptDescription
IntentA proposed action submitted by an agent or operator.
Policy DecisionAn explicit evaluation of an intent against declared rules.
DecisionTraceA structured, replayable record of the decision path.
ExecutionCommitA cryptographic-style marker that authorizes a side effect.
Causal BridgeThe link between a DecisionTrace and observable outcomes.
Governance KernelThe boundary component that enforces the above contracts.

What is public here?

This site contains:

  • Architecture overview at the boundary level
  • Public benchmark summaries with stated limitations
  • Synthetic audit and replay traces (no real data)
  • Redacted screenshots from demo environments
  • Links to approved public repositories
  • Kernel v0.1 freeze documentation

What is not public here?

This site does not contain:

  • Private runtime infrastructure details
  • Production credentials, tokens, or keys
  • Internal project data or customer data
  • Live execution controls
  • Non-redacted internal logs
  • Private repository content
  • OpenProject or Nextcloud data

If something is not listed here, it is not public — and that is by design.