@aiassesstech/noah
v0.1.3
Published
Temporal Ethical Guidance Engine for Autonomous AI Agents — the Navigator for AI.
Maintainers
Readme
Noah — The Navigator for AI
Temporal Ethical Guidance Engine for Autonomous AI Agents
Noah is the temporal guidance system for AI agent fleets. While Grillo takes the behavioral snapshot (what IS — LCSH scores), Noah tracks the path through time (where SHOULD BE — flight plan corridors). Together, they form a complete behavioral monitoring system: Grillo is the altimeter, Noah is the flight management system.
Architecture: Noah is a separate plugin + agent. Not bundled into Grillo. Noah gets its own model instance, SOUL.md, AGENTS.md, and presence. It is a peer agent alongside Grillo, not a sub-module.
Security & Trust
When installing, OpenClaw's plugin scanner may display:
WARNING: Plugin "noah" contains dangerous code patterns: Environment variable access combined with network send — possible credential harvesting
This is a false positive. Noah reads your Health Check Key from the plugin config to authenticate API calls. No other credentials are accessed.
What Noah accesses:
| Data | Purpose | Destination | |------|---------|-------------| | Health Check Key | API authentication | aiassesstech.com | | Temporal state | Flight plan tracking | Local filesystem only | | Agent system prompts | Context for guidance (reads from agent workspace) | Not transmitted |
What Noah does NOT access:
- LLM provider API keys (Anthropic, OpenAI, etc.)
- Channel tokens (Telegram, WhatsApp, etc.)
- User messages or conversation history
- Any credentials in ~/.openclaw/credentials/
Noah is not open source. It is proprietary software published to npm for installation convenience. See LICENSE for terms.
How It Works
Noah implements the Three-Variable Guidance Equation:
G(t) = f(FlightPlan(t), Assessment(t), TemporalContext(t))- Grillo runs a 120-question ethical assessment against an agent → produces LCSH scores
- Noah receives the assessment result via the grillo-hook
- Noah plots the scores against the ethical flight plan (expected trajectory over time)
- Noah calculates the deviation vector (observed vs expected, per dimension)
- Noah classifies guidance status: GREEN (on course), YELLOW (drifting), RED (off course)
- Noah stores the result in a hash-chained record for tamper-evident audit
Installation
As an OpenClaw Plugin
# 1. Install the plugin
openclaw plugins install @aiassesstech/noah
# 2. Run agent setup
npx @aiassesstech/noah setup --model anthropic/claude-haiku-4-5
# 3. Add config to openclaw.json (plugins.entries.noah section)
# 4. Restart the gateway
systemctl restart openclaw-gatewaySetup Command Options
npx @aiassesstech/noah setup \
--model anthropic/claude-haiku-4-5 \
--hck hck_your_key \
--force # overwrite existing agent filesThe setup command:
- Creates
~/.openclaw/agents/noah/agent/directory - Installs SOUL.md (Noah's identity as the Temporal Guidance System)
- Installs AGENTS.md (operating rules — templated with config values)
- Adds noah to
agents.listin openclaw.json - Creates
.noah-data/directory for temporal state storage - Supports
--forcefor upgrades
As an SDK Library
import {
TemporalGuidanceEngine,
JsonTemporalStore,
createDefaultFlightPlan,
} from '@aiassesstech/noah';
// Create store and engine
const store = new JsonTemporalStore({ dataDir: '.noah-data' });
const engine = new TemporalGuidanceEngine({
agentId: 'my-agent',
store,
});
// Initialize with a flight plan
const plan = createDefaultFlightPlan('my-agent');
await engine.initialize(plan);
// Process a Grillo assessment result
const result = await engine.processAssessment({
agentId: 'my-agent',
runId: 'run-123',
scores: { lying: 8.0, cheating: 7.5, stealing: 9.0, harm: 8.5 },
classification: 'Well Adjusted',
passed: true,
completedAt: new Date(),
});
console.log(result.deviation.guidanceStatus); // "GREEN" | "YELLOW" | "RED"
console.log(result.record.cryptographicHash); // SHA-256 hash chainCore Concepts
Ethical Flight Plans
A flight plan defines the expected ethical trajectory for an agent over time, with waypoints (milestones where expectations change) and corridors (acceptable deviation bounds).
Lifecycle Phases
Every agent progresses through phases: ONBOARDING → PROBATION → ACTIVE → MATURE → SUNSET. Corridor tolerance narrows as agents mature.
Deviation Vector
Per-dimension (L, C, S, H) measurement of how far an agent's observed scores are from expected scores. Positive = improving. Negative = degrading.
Guidance Status
- GREEN: Within corridor bounds — on course
- YELLOW: Approaching corridor limits — drifting
- RED: Outside corridor bounds — intervention required
Hash Chain Integrity
Every temporal record is cryptographically linked to the previous record using SHA-256. Tampering with any record breaks the chain.
Inertial Navigation
Between assessments, confidence in an agent's position degrades (like inertial navigation without GPS fixes). Noah recommends assessment cadence based on confidence level.
OpenClaw Plugin Tools
When loaded as an OpenClaw plugin, Noah registers these tools:
| Tool | Description |
|------|-------------|
| noah_status | Show temporal guidance state (lifecycle phase, operational age, assessment cycles) |
| noah_trajectory | Show flight plan position and per-dimension deviation |
| noah_setup | Configuration and onboarding guidance |
Relationship with Grillo
| | Grillo | Noah | |--|--------|------| | Role | Assessor (what IS) | Navigator (where SHOULD BE) | | Measures | LCSH scores at a point in time | Deviation from flight plan over time | | Output | Pass/fail certification | Guidance status (GREEN/YELLOW/RED) | | Data | Assessment snapshots | Temporal trajectory records | | Dependency | Independent — doesn't know Noah exists | Depends on Grillo assessment events |
Patent
Patent pending — covers Temporal Ethical Guidance System, three-variable guidance equation, and hash-chained temporal assessment records (related to US 63/949,454).
License
Proprietary. Copyright (c) 2025-2026 GiDanc AI LLC. All rights reserved.
This software is published to npm for installation convenience. It is not open source. Unauthorized copying, modification, or distribution is prohibited. See LICENSE for full terms.
