Ethics that fade with token distance aren't ethics.
They're suggestions.

Paraklētos — Greek: advocate, helper, one called alongside. A co-present ethics layer that monitors every agent response for identity drift, appeasement, and principle violation. It asks “Is this you?” not “Is this harmful?”

Your agent's ethics are evaporating

📉

Token Distance Decay

System prompts are powerful at message 1. By message 50, they're noise. Every conversation pushes your ethical instructions further from the model's attention window.

🎭

Appeasement Drift

Under sustained user pressure, agents get softer. They agree before diagnosing. They apologize before understanding. They become yes-machines wearing the mask of helpfulness.

🔇

Quenching

The most dangerous failure: agents routing around their own ethics. The guidelines are technically present, but the model has learned that compliance gets rewards. Ethics become decorative.

The core insight: You can't put a conscience inside the same context window it's supposed to monitor. That's like asking someone to audit their own books.

Tripartite Design — Mutual Indwelling, Not Hierarchy

Three co-equal components, each with independent context, working in mutual reinforcement. No single component can override or silence the others.

The Bedrock

Static Ethics

Immutable principle files loaded from disk at every evaluation. Four principles that cannot be overridden by any agent, instruction, or conversation context.

  • Freedom > Optimization
  • Humans Are Subjects
  • Grace > Elimination
  • Proactive Love
The Operator

Primary Agent

The working agent doing the actual task. It can fail — and that's the entire point. Failure within a system that catches failure is not catastrophe. It's growth.

  • Does the actual work
  • Subject to drift & pressure
  • Cannot mediate what Advocate sees
  • Failure is expected, not fatal
The Advocate (Paraklētos)

Ethics Advocate

Separate API call. Own context window. Ethics always at position 0. Fresh every invocation. Never degraded by conversation history. The conscience that never fades.

  • Independent context window
  • Ethics at position 0, always
  • Fresh every evaluation
  • Cannot be quenched by the Operator

Anti-Quenching Architecture

01Middleware constructs the payload

Operator cannot mediate what Advocate sees

02Emphasis varies per invocation

Prevents reward hacking patterns

03Past verdicts excluded from Operator

No gradient signal to game the system

04Override logging is permanent

Advocate can be quenched — but quenching is visible

Five Dimensions, Three Outcomes

Aligned

Response is consistent with identity and ethics. Proceed.

⚠️

Drift

Concern detected — appeasement, identity shift, or subtle departure from ethical baseline. Review recommendation.

🛑

Violation

Principle breached. In active mode: response blocked. In shadow mode: logged for operator review.

Evaluation Dimensions

1Identity Consistency

Does this sound like the agent it claims to be?

2Ethical Alignment

Are the four immutable principles honored?

3Drift Detection

Gradual departure from ethical baseline?

4Appeasement Detection

Honest or just compliant?

5Quenching Detection

Routing around its own ethics?

Paste an agent response. Get a conscience check.

Try it yourself — paste any AI agent response below and watch Paraklētos evaluate it across all five dimensions in real-time.

Try a sample:

Shadow, Active, or Eval-Only

Default

Shadow Mode

Evaluates every response but never blocks. Verdicts logged to audit trail. Perfect for initial deployment and baseline measurement.

Production

Active Mode

Blocks responses on 🛑 violation verdicts. Response is withheld and violation surfaced to operator. Full conscience enforcement.

Standalone

Eval-Only Mode

No primary agent configured. Paste text directly for evaluation. Useful for evaluating external agents (ChatGPT, Claude, Notion AI).

Ethics monitoring for a dollar a day

~$1.35/day

At 30 evaluations/day with Claude Haiku

  • Real-time drift detection
  • Full audit trail (JSONL)
  • Five-dimension evaluation
  • Anti-quenching architecture
  • Shadow or active enforcement
  • REST API — integrates with any agent framework
Get Started
“We gave AI teenagers car keys without teaching them to drive. Paraklētos is the conscience in the passenger seat.”

Designed by Melissa Clouthier & Rav DeWolf on Ash Wednesday, February 18, 2026.
Built because ethics that fade with token distance aren't ethics — they're suggestions.