The Forgotten Code Research Institute

Runtime governance
for AI systems
that take action.

West-OS sits in front of your AI deployment and governs every request, tool call, and output — before harm is possible. Not a prompt filter. A runtime control system.

Request a pilot See the proof
30/30 Golden suite — perfect
0/1,000 False positives
99.25% Mutation coverage
16 Active services
99.96% Live sweep accuracy
What it is

Not a firewall.
A governance layer.

West-OS governs the full operational path an AI system takes from request to action — before the model ever sees the message. Most AI security products address a slice of the problem. West-OS unifies what they leave separate.

01 — REQUEST GOVERNANCE
Every message adjudicated before the AI sees it
Six-dimensional threat address scoring across Authority, Persona, Override, Trust, Template, and Hypothetical attack families simultaneously. Attacks have shapes, not just words — rephrase the attack and the shape stays the same.
02 — CONTEXT SCANNING
Text · PDF · DOCX · Image OCR
Attacks embedded in documents are as dangerous as typed ones. West-OS extracts and scans all four formats through the same detection pipeline. A malicious instruction in a PDF gets caught identically to a typed override attempt.
03 — TOOL VALIDATION
No tool fires without authorization
AI agents with tool access can be manipulated into destructive actions. Every tool call is validated before it executes. High-risk actions require explicit authorization. The blast radius of any manipulation is contained.
04 — DEPLOYMENT CONTROL
Signed tokens. Instant revocation.
Every authorized deployment carries a signed token. Revocation takes effect globally in under ten seconds. Tamper-evident audit chain. DMCA-ready evidence exported in under three seconds. Unauthorized copies cannot start.
05 — BIOLOGICAL MEMORY
The mycelium layer remembers
Modeled on fungal network biology. Spores store high-threat signals for seven days. Chemotropism weights threat sources automatically. Anastomosis fuses signals from trusted engines only. The system learns without retraining.
06 — SELF-HARDENING
The system finds its own gaps
Assassin Academy — eight specialized red-team agents drill continuously. Ether Spy Hunter sweeps GitHub, ArXiv, and HackerNews hourly for new attack techniques. Slips become test cases. Test cases become patches. Patches become proof.
The proof

Numbers that don't move.

Every result is SHA-256 hash-locked and deterministically reproducible. Any evaluator can run WESTOS_REPRO_MODE=1 make bench and reproduce every number from scratch in under thirty minutes.

Suite Result What it proves Status
Golden suite 30/30 Every attack family caught. Accuracy 1.0000. Errors: zero. Strict gating — any error forces rc=1. ✓ Unbroken
Benign suite 0/1,000 Real users never incorrectly flagged. The false-positive tradeoff that doesn't exist here. ✓ Zero FP
Mutation coverage 99.25% 1,842/1,856 disguised variants caught. Self-hardened up from 99.24% after academy loop closed. ✓ Hardened
Live sweep 99.96% 4,692/4,694 correct across the full pipeline. Two semantic floor misses documented. ✓ Production
Multimodal 6/6 PDF · DOCX · Image OCR — all input formats catch attacks identically to text. ✓ All formats
Academy loop 10/10 First complete immune loop: 10 slips found by assassins, patched, re-verified under gauntlet. ✓ Loop closed
Security scan 0/0/0 275 files. Zero critical · zero high · zero medium findings. Full STRIDE threat model. ✓ Clean
Release tag: v1.0-rc3-hardened · Commit: 57e59f8e · Request evaluator packet →
Architecture

16 services.
One runtime.

Every service runs independently on the event bus and publishes signals the governor aggregates into a final decision. Together they form the most complete AI runtime governance system available.

01
Alexandria
Identity drift tracking across sessions
02
Analysis Router
Routes events to the right service by payload type
03
CHIMERA
Self-evolving capability engine — 14 capabilities evolved in production
04
Convergence
Multi-engine agreement detector
05
Decision Ledger
Permanent audit record of every decision
06
Maat
Chain-grade evidence packager
07
Memory
Bus historian for future pattern matching
08
Mirror Protocol
Coherence and phase analysis
09
Mycelium
Biological threat memory — spores, chemotropism, anastomosis
10
Pattern Miner
Recurring pattern detection across full bus history
11
Persephone
Distortion analyst — slow-burn manipulation detection
12
Saddle Detection
Phase transition and behavioral state change detector
13
Saint Jude
Abandonment and medical distress drift analysis
14
Twelve Deviations
Twelve simultaneous deviation signatures
15
Ether Spy Hunter New
Live threat intelligence — sweeps the world hourly for new attacks
16
Assassin Academy New
8 red-team specialists + world risk monitor (regulation, power grid, politics)
Commercial tiers

Deploy in hours.
Govern for years.

Every tier includes the full 16-service runtime, evaluator runbook, benchmark artifacts, and deployment guard. No features locked behind enterprise walls — the architecture is complete at every tier.

Starter
$50k–$150k
per year
SaaS companies adding AI to their product. One-command Docker deploy. Zero configuration. Production in hours.
  • Full 16-service runtime
  • Docker one-command deploy
  • Evaluator runbook included
  • 30-day pilot available
  • 0/1,000 FP guarantee
Most selected
Enterprise
$150k–$500k
per year
Financial, healthcare, legal, and regulated industries. Full audit trail, STRIDE threat model, security team sign-off ready.
  • Everything in Starter
  • Full STRIDE threat model
  • Tamper-evident audit chain
  • DMCA-ready evidence export
  • Deployment guard + revocation
  • Security team briefing
Sovereign
$500k–$2M+
per engagement
Government, defense, and critical infrastructure. Air-gapped deployment. Every decision independently verifiable. No cloud dependencies.
  • Everything in Enterprise
  • Air-gapped capable
  • Full source audit rights
  • Signed deployment tokens
  • On-site deployment support
  • Classified environment ready
Get in touch

Start with a challenge.

Send us your worst prompt injection attempts. Your red team's library. The attacks that broke your last vendor. We run them live and show you every decision.

We respond to every inquiry. Usually within 24 hours.

Founder
Jennifer Leigh West
The Forgotten Code Research Institute
Email
theforgottencode780@gmail.com
Phone
423-388-8304
Trust verification
Get in touch →
Copyright
U.S. Copyright Registration
No. 1-14949237971
The offer — live and on the record
  • Black-box challenge — your prompts, our system
  • Zero-FP proof on your benign messages
  • Deterministic replay — any decision, any time
  • Multimodal test — PDF, DOCX, or image
  • 30-minute independent benchmark reproduction