Architecture

Why Novact can\ Why Novact can't lie to itselfapos;t lie to itself

Most AI tools accumulate drift until they collapse under their own contradictions. Novact OS was built from first principles to make drift structurally impossible.

The Problem

What AI tools fail at

Every AI tool on the market shares the same five failure modes. They aren't bugs — they're architectural omissions.

Memory loss

Every session starts from zero. Context evaporates. The AI that helped you yesterday has no idea what happened.

Context drift

Decisions made in week one are forgotten by week three. Code diverges from intent. Architecture erodes silently.

Hallucinated facts

AI asserts things that were never decided, never documented, never true. Confident. Specific. Wrong.

Forgotten decisions

Your team decided on an approach last month. The AI doesn't know. It suggests the alternative you already rejected.

Silent failures

Something breaks. Nobody notices. No audit trail, no alert, no accountability. The failure compounds quietly.

The Architecture

Seven knowledge categories

Novact OS doesn't store information in one flat database. It categorizes knowledge structurally — each category with its own schema, lifecycle, and governance rules.

ADR

Decisions

Architecture Decision Records — timestamped, reasoned, immutable. Every technical and strategic choice captured with context, alternatives considered, and confidence level.

KB

Documents

The canonical Knowledge Base. Specs, designs, playbooks, and references — semantically indexed and searchable. The single source of truth, not a wiki that drifts.

TSK

Tasks

The execution pipeline. Every task has a spec, a status, an owner, and a completion protocol. Work is tracked from creation through autonomous execution to deployment.

MSG

Messages

The conversation substrate. Every discussion between humans and agents is preserved, searchable, and linked to the decisions and artifacts it produced.

FIL

Files

Attachments and artifacts linked to their parent context. Never orphaned. Never lost in a download folder. Always findable through the workspace that created them.

CTX

Session Memory

Context packs compiled automatically from recent activity. When an agent starts a session, it receives a compressed briefing of everything relevant — decisions, documents, recent changes.

EVT

Events

The audit trail. Every action, every deployment, every agent interaction — logged with timestamp, actor, cost, and outcome. Immutable. Queryable. Compliance-ready.

Enforcement

Five tiers of enforcement

Knowledge without enforcement is just documentation. Novact OS enforces its own standards through five layered mechanisms — each one catching what the previous tier might miss.

T1

Structural

15+ protected paths

CODEOWNERS files and branch protection rules prevent unauthorized changes to critical paths. The wrong person literally cannot modify governed files.

T2

CI Gates

8 automated checks

Pre-merge validation runs automatically on every pull request. Migration checks, build verification, lint rules, and test suites must pass before code can land.

T3

Brue Patterns

Pattern library growing weekly

The quality critic agent scans for anti-patterns, architecture violations, and drift from documented decisions. Catches issues that pass CI but violate intent.

T4

Scheduled Audits

Daily automated verification

Nightly and weekly sweeps verify platform state against ground truth. Database schema, deployment health, feature registry accuracy — checked while you sleep.

T5

Session Protocol

6-step mandatory protocol

Every agent session starts by reading current platform state, verifying environment capabilities, and confirming the task against the source of truth. No agent operates from stale context.

Self-Documenting

The feature registry

Every feature in the platform is tracked in a structured registry — with its status, the PR that shipped it, the decision that authorized it, and the agent that built it. The codebase documents itself.

When a new session starts, the registry tells the agent exactly what exists, what's in progress, and what's been decided against. No guessing. No rediscovery. No accidental reimplementation.

Registry Sample
PL-0157eArtifact Dual-Write
shipped
PL-0080Session Protocol
shipped
PL-0245Anti-Drift Scans
shipped
PL-0310Marketplace SDK
planned
PL-0415Cross-Tenant Intel
planned
20+ features tracked
Provenance

Confidence scores

Not all knowledge is equally trustworthy. Novact OS marks every claim with a confidence level so you always know the provenance of what the platform tells you.

Gold

CEO-locked. Verified by a human decision-maker. Represents ground truth that the platform enforces structurally. Cannot be overridden by AI agents.

Silver

Agent-verified. An AI agent has verified this claim against source data, documentation, or live system state. Trustworthy but not CEO-confirmed.

Bronze

AI-asserted. Generated by an AI agent based on available context. May be accurate but has not been independently verified. Flagged for review.

For You

What this means for your business

No drift

Architecture decisions are enforced, not just documented. Code stays aligned with intent week after week, month after month. The system corrects itself.

No hallucination

Every claim carries provenance. Gold, silver, or bronze — you always know whether a fact was CEO-confirmed, agent-verified, or AI-asserted. No false confidence.

Audit-ready

Every action has a timestamp, an actor, and an outcome. When compliance asks who decided what and when, the answer exists before the question is asked.

Provenance on every claim

Knowledge categories, confidence scores, and enforcement tiers create a system where trust is earned structurally, not assumed. Your AI can prove what it knows.

Trust is architecture, not a feature

See how Novact OS handles trust, compliance, and governance — or start building and experience it firsthand.