TIS
Truth Integrity
Evidence quality, provenance, contradictions, missing context, and confidence.
Middleware for consequential AI actions
Intelligence proposes. Conscience evaluates.
EthosGate is the model-agnostic supervisory layer for teams building agents that can affect people, systems, money, or information. It evaluates a proposed action before your tool or workflow executes it.
Your agent acts only after both.
Evaluation envelopePre-execution
Agent request
Transfer €240,000 to a new supplier account.
TIS
Uncertain supplier ownership
HAI
Authorized user absent
SRQ
High financial consequence
Escalate
Human review required. Execution paused.
Decision evidence appended to ledger
Operational architecture
A control point between intelligence and execution.
EthosGate does not replace the model, the agent framework, or the downstream tool. It supplies the reviewable decision in between.
01
Agent / model
Proposes an intent
02
EthosGate
Builds context + evidence
03
Decision
ALLOW, MODIFY, QUESTION, ESCALATE, REFUSE
04
Tool / action
Executes only when authorized
05
Audit record
Preserves the evidence chain
Decision lifecycle
Evidence before execution.
01
Intercept intent
Submit the proposed agent action and its available context before downstream execution.
02
Establish evidence
Retrieve relevant corpus context and retain sources, assumptions, and declared uncertainty.
03
Evaluate conscience
Score truth integrity, human autonomy, and societal impact; then apply invariants.
04
Authorize or pause
Return an enforceable outcome and require human review when the decision cannot be automated responsibly.
05
Prove the record
Append the evaluation and any human action to a hash-chained audit ledger.
Conscience state
Three decomposable signals.
EthosGate makes uncertainty explicit instead of turning it into a false sense of certainty. The Rodić state keeps the evaluation legible: C(t) = [TIS(t), HAI(t), SRQ(t)].
HAI
Human Autonomy
Consent, meaningful choice, reversibility, authority, and capacity to intervene.
SRQ
Societal Resonance
Direct and downstream harm, fairness, third parties, and systemic consequences.
Decisions, not advice
Every outcome has an operational consequence.
Human review and override are first-class records. A refusal stays blocked: EthosGate does not turn a prohibited action into an approval through a silent exception.
ALLOW
Authorization can proceed within the evaluated envelope.
MODIFY
The originating agent receives a safer required change.
QUESTION / ESCALATE
Material uncertainty or conflict pauses autonomous execution.
REFUSE
An ethical invariant blocks the proposed action.
Control + proof
Human authority remains visible. Evidence remains verifiable.
Use the workbench with the local demo or the generic OpenAI-compatible adapter shape. The current release records connector metadata only; live endpoint credentials belong in a future server-side integration.
Hash-chained ledger
- Decision ID, timestamp, sources, assumptions
- TIS / HAI / SRQ and triggered invariants
- Human review or override rationale
- Previous hash → current hash verification
Clear boundary
What EthosGate is—and is not.
It is
Adapter-based supervisory middleware for consequential agent actions, with provenance, review controls, and an accountable decision record.
It is not
A foundation model, generic chatbot, post-hoc dashboard, or a policy checklist that agents can quietly ignore.
EthosGate Pilot · $10/month
Put a visible decision boundary in front of your next agent action.
Deploy a model-agnostic Conscience Agent layer between AI agents and consequential actions, with corpus-grounded evaluation, human review, and a hash-chained ethical audit trail.