Recover your context. Control your agents.
Memory to recover context and authorisation controls for agent actions. Review what has been tested and its limits.
Your existing AI tools and agents (Claude Code, Cursor, MCP servers, dev scripts, or local pipelines). Zero changes to your models.
CORTEX Memory (your AIs remember context) + BABYLON Guard Firewall (blocks AI before it deletes or breaks your code) + NEMESIS immutable audit log.
A single Rust binary. Everything runs on your machine (Local-First). No sending code to the cloud, no third-party external databases, and no forced subscriptions.
Your AI acts. Your business is accountable.
The EU AI Act puts use, risk and responsibility in focus. Understand which obligations may apply and the controls BABYLON documents to support your work.
- Human oversightWho authorises, who can intervene and which effect is observed.
- TraceabilityConnect decisions, authorisations and outcomes with evidence.
- Risk managementAssess the actual use case, its limits and accountable owners.
Find what you forgot.
Recover valuable ideas buried across years of conversations, notes, and projects.
"Agents should require explicit authority before modifying or deleting files."
Static demonstration. In production, resolves directly to the immutable SourceRecord sealed by SHA3-256.
Static demonstration of destructive mutation interception prior to filesystem commit.
Let AI work. But you keep the wheel.
BABYLON documents authorisation controls for actions on protected resources. Coverage depends on integration and configuration; review the evidence and paths not yet verified.
Don't ask for promises. Demand mathematical proof.
Review adversarial test results, observed states and documented divergences. These define the scope of the evidence, not a global security guarantee.
Proven in Code
We mathematically prove that the engine security rules have no logical flaws.
Concurrent Shielding
Immune to timing attacks (race conditions) and rapid file manipulation.
Zero Corruption
If the computer crashes suddenly, the system recovers instantly without losing a single byte of evidence.
Anti-Deception
AI cannot deceive the system by camouflaging destructive commands under another name.
100% Auditable
Any engineer can clone our test harness, run the tests locally, and verify the truth.
Terminal Observability with babylon top
Sample projection based on a static capture of the TUI executed on a local test harness (Darwin arm64 on APFS):
Empirical demarcation notice: This projection is a static capture of the babylon top TUI archetype executed on a local test harness (macOS Darwin arm64 on APFS). It is not a live remote stream. In the current release, all 6 invariants are cataloged as STALE pending automated continuous pipeline attestation. Review verified bounds in Assurance & Limits.
Past, Present, and Future
How the three domains coordinate around the agent lifecycle:
CORTEX
CORTEX · Recover context and decisions
BABYLON
BABYLON · Delimit and govern effects
NEMESIS
NEMESIS · Popperian Falsification
The Silicon Machinery Beneath the Product
No stochastic hand-waving: every interaction is bound to deterministic hardware primitives, cryptography, and systems theory:
C5-REAL Substrate
Landauer dissipation bounds, Chentsov information metric, and asymptotic compression without operational friction.
WORM Ledger
Single-writer SQLite WAL with Merkle Mountain Ranges enabling O(log N) inclusion proofs.
Effect Broker & 7-Tuple
Every mutation evaluates ⟨Authority, Capability, Policy, Resource, Effect, Context, Epoch⟩ fail-closed.
Hardware Touch ID Gate
Local Apple Silicon biometric friction via NIST P-256 Secure Enclave with zero reuse duration.
Popperian Falsification
NEMESIS challenges security assertions under adverse fault models. Attestations bind evidence, not ontic truth.
Decoupled Causal Memory
Axiom O1: Source ≠ Inference. Primary records remain immutable under cryptographic SHA3-256 seals.