Every admitted fact receives timestamp, type, tenant scope, provenance, and ledger continuity metadata.
CORTEX Persist / verifiable memory layer
local-first
hash chained
tenant scoped
Memorymust proveit happened.
CORTEX turns agent memory into evidence: every admitted fact carries provenance, tenant scope, timestamps, guard status, and cryptographic continuity.
418
sample tx classified
4
evidence outputs
3
review gaps flagged
1
sealed audit manifest
Designed to sit underneath the tools teams already use
Claude Code
Cursor
LangChain
CrewAI
AutoGen
Custom agents
Regulatory pressure
Article 12 needs evidence, not screenshots.
High-risk AI systems in the EU need automatic logging, traceability, and integrity controls. CORTEX is not legal advice and does not make a system compliant by itself. It gives your team the technical record needed to prove what an agent knew, decided, and wrote.
Potential exposure
up to EUR 30M or 6% of worldwide annual turnover
For qualifying EU AI Act violations, depending on classification and facts.
Hash chains and Merkle checkpoints make record mutation detectable instead of silently accepted.
Facts, votes, guards, and verification state can be traced back to the operation that created them.
WBFT voting and disputed-state handling keep single-agent memory writes from becoming unreviewed truth.
Operating model
From agent output to evidence chain.
CORTEX does not replace your model, framework, or vector store. It adds the deterministic trust boundary between raw output and persisted memory.
Propose
An agent proposes a fact, decision, observation, or runtime state change.
Validate
Guards enforce schema, taint, tenant scope, secret redaction, and deterministic admission rules.
Persist
Accepted state is written transactionally with ledger continuity and explicit failure semantics.
Verify
Teams can inspect, replay, export, and verify the chain when incidents or audits happen.
What buyers get
An evidence pack your security team can inspect.
For serious teams, the value is not another dashboard. The value is a defensible record of what was admitted, rejected, verified, and scoped.
Artifact 01
Ledger continuity report
Shows chain status, checkpoint coverage, schema version, and the exact verification command used to reproduce the result.
cortex trust-ledger verify --tenant acme
Artifact 02
Guard rejection log
Explains why a proposed memory write failed, including schema, taint, tenant, secret, and validation boundaries.
Artifact 03
Tenant scope map
Documents which facts belong to which tenant so public reads do not accidentally mix customer or workspace memory.
Artifact 04
Incident replay context
Reconstructs what the agent knew at a point in time without relying on screenshots, chat transcripts, or vendor console history.
Agente especializado
CriptoFiscal AEAT convierte actividad on-chain en un expediente defendible.
Una capa fiscal-forense para reconstruir operaciones cripto/NFT, valorar en euros, documentar fuentes y entregar evidencia revisable por asesores antes de cualquier declaración o requerimiento.
De wallets dispersas a una línea temporal verificable.
El agente normaliza CSVs de exchanges, movimientos entre wallets propias, DEX swaps, bridges, mints, burns, royalties y ventas NFT. Cada entrada conserva hash, fecha/hora, fuente de precio, umbral de riesgo y estado de evidencia.
Arranca con una auditoría simulada para fijar umbrales de riesgo antes de automatizar extracción.
Ingiere wallets, CSVs, explorers y marketplaces sin mezclar transfers propios con transmisiones reales.
Separa compra, venta, permuta, airdrop, staking, mint, burn, bridge, royalty, comisión y custodia.
Entrega valoración en euros, incidencias, trazabilidad de precios y anexos para revisión profesional.
Límite operativo: CriptoFiscal AEAT no inventa titularidades, precios ni criterios fiscales. Contrasta obligaciones con fuentes oficiales vigentes y mantiene claro qué es hecho verificado, inferencia razonable o dato pendiente.
Where it fits
Built for teams where memory becomes liability.
Regulated AI products
Maintain traceable decision memory for fintech, health, insurance, legal, and public-sector AI workflows.
Multi-agent engineering
Keep architecture decisions, guard outcomes, and handoffs consistent across long-running agent swarms.
Security-sensitive deployments
Run local-first memory with tenant scoping and ingress checks before facts enter durable state.
Core capabilities
Memory infrastructure with a trust contract.
Most memory products optimize recall. CORTEX optimizes whether a recalled fact should be trusted.
Immutable ledger
Facts are appended to a SHA-256 hash chain, with verification commands for continuity checks.
Tenant boundaries
Public read and write paths are designed around explicit tenant scope to avoid cross-customer memory bleed.
Guarded admission
Validation failures stay failures. Unverified generative output does not become durable fact by accident.
Hybrid retrieval
Semantic and full-text recall can work together while preserving provenance and verification state.
WBFT consensus
Multiple agents can vote on facts with weighted Byzantine fault-tolerant agreement semantics.
Local-first runtime
Start with SQLite on your machine. Add cloud, sync, or enterprise deployment only when the risk model allows it.
Full API surface
CORTEX is not just a landing page. It is an operator API.
The backend exposes a broad FastAPI, CLI, SDK, MCP, and daemon surface. The product page should make that clear before buyers assume this is only a checkout wrapper around a ledger.
/v1/facts
Admit tenant-scoped memory only after schema, taint, provenance, and ledger checks pass.
{
"tenant_id": "acme",
"type": "decision",
"taint": "CORTEX-TAINT"
}/v1/search
Retrieve semantic and lexical context without crossing tenant boundaries.
query=incident replay
trust-ledger
Replay hash continuity, checkpoints, and evidence manifests before export.
cortex trust-ledger verify
agent handoff
Expose durable context to operators, SDK clients, and agent tools with explicit scope.
memory.passport.read
55+
documented API endpoints
90+
operator CLI commands
13
daemon monitors
4
access surfaces: API, CLI, SDK, MCP
Store, recall, edit, deprecate, and ghost facts
Durable facts are not generic notes. They carry type, project, tenant, timestamps, provenance, and trust metadata.
POST /v1/facts
cortex store
Semantic + lexical retrieval
Use vector search, FTS5, graph-aware recall, temporal filters, and ranked context assembly for agents.
POST /v1/search
GET /v1/search
Verify hash continuity and checkpoints
Ledger verification, Merkle roots, snapshots, and exportable evidence turn memory into an auditable state chain.
cortex trust-ledger verify
cortex export
Multi-agent voting and reputation
Agents can vote on facts with quorum, weighted reputation, disputed states, and consensus audit trails.
cortex vote_v2
consensus helpers
Point-in-time reconstruction
Export project memory, navigate temporal state, reconstruct incidents, and produce auditor-friendly context packs.
GET /v1/projects/{project}/export
cortex timeline
RBAC, API keys, tenant scope
Roles, permissions, API keys, rate limits, and explicit tenant boundaries protect public read/write surfaces.
Authorization: Bearer ctx_...
tenant_id scoped reads
Health, metrics, watchdogs
Expose health checks, Prometheus metrics, observatory state, daemon status, and operational signals.
GET /health
GET /metrics
GET /v1/daemon/status
Stripe, Langbase, MCP, SDKs
Use server-side billing routes, async clients, MCP tools, language integrations, and hosted API surfaces where needed.
POST /v1/stripe/checkout
POST /v1/langbase/search
Agéntica layer
A control plane for societies of agents.
CORTEX also includes an agéntica stack: registered agents, heartbeat supervision, swarm missions, consensus resolution, handoff state, and self-improvement loops. The memory layer is only the substrate; the agent layer is the operating system on top.
Supervisor
MemoryAgent
SecurityAgent
VerificationAgent
HandoffAgent
NightshiftAgent
OmegaPrime
From single-agent memory to coordinated agency.
Agents can register, emit heartbeats, receive messages, persist working memory, vote on facts, and escalate conflicts. Swarms can run isolated worktrees, budget execution, verify risky changes, and record mission results back into the ledger.
Agent registry
Message bus
Heartbeat supervision
Swarm missions
WBFT arbitration
Handoff state
Agent lifecycle
Register agents, monitor liveness, restart supervised tasks, and keep working memory attached to agent identity.
POST /v1/agents
GET /v1/agents
Swarm orchestration
Coordinate parallel agents, isolated worktrees, mission queues, budget controls, and high-risk verification gates.
GET /v1/swarm/status
POST /v1/missions
Mejoralo engine
Scan, record improvement sessions, inspect history, and run ship gates before autonomous changes become release candidates.
POST /v1/mejoralo/scan
POST /v1/mejoralo/ship-gate
Conflict resolution
Use weighted votes, Byzantine consensus, elder-council style escalation, and audit trails for disputed agent decisions.
consensus_score
vote ledger
Investigación intelligence
Deep research that becomes durable knowledge.
CORTEX can run research as an evidence pipeline: discover gaps, extract sources, generate audits, synthesize findings, and persist only traceable claims. This turns investigation from a chat transcript into a reusable intelligence asset.
Discover
Knowledge Radar scans curated targets, unresolved ghosts, semantic gaps, and operational signals.
Acquire
Scraper-Ω and bounded file/API tools collect source material without mixing raw evidence with verified facts.
Analyze
Oracle, Mejoralo, swarm formations, and static-analysis gates produce dense reports and failure signatures.
Crystallize
NightShift and ledgered memory convert high-confidence findings into reusable, timestamped knowledge.
Competitive analysis, architecture audits, security research, and scientific synthesis.
Turn competitive analysis, architecture reviews, scientific notes, and security findings into reusable intelligence with source trails, timestamps, and explicit confidence boundaries.
Fast B2B audit reports
Run targeted audits against companies, repos, APIs, or architecture surfaces and store the report back into CORTEX.
POST /v1/oracle/audit
One endpoint, full picture
Aggregate system status, daemon data, trust posture, and evolution/swarm signals into a single operator view.
GET /v1/observatory
Find what the system does not know
Detect unresolved ghosts, semantic gaps, and queued targets so the system can prioritize research automatically.
knowledge_gap
Autonomous knowledge generation
Background cycles discover targets, build plans, forge knowledge crystals, and persist cycle reports.
nightshift_cycle
Especialidad fiscal blockchain + NFT
De wallets a expediente: evidencia técnica para Hacienda.
CORTEX puede convertir actividad onchain dispersa en un paquete verificable para revisión fiscal: wallets, exchanges, swaps, NFTs, royalties, coste de adquisición, gaps de custodia y trazabilidad por hash.
Una historia fiscal reconstruible, no una hoja de cálculo frágil.
El objetivo no es sustituir al asesor fiscal. Es entregar evidencia técnica ordenada, reproducible y sellada para que un profesional pueda revisar la posición con menos fricción y más confianza.
$ cortex audit crypto --wallet 0x7a... --year 2025
classified: buys, sells, swaps, nft-sales, royalties
gaps: 3 require manual evidence
export: pdf + csv + hash manifestReconstrucción de wallets, contratos, exchanges, bridges y direcciones relacionadas.
Soporte para coste de adquisición, precio de transmisión, fees, timestamps y moneda base.
Ventas, compras, royalties, marketplace fees, metadata y tx hashes verificables.
Marca operaciones incompletas, fuentes ausentes, movimientos internos y puntos que requieren criterio fiscal.
Boundary: CORTEX genera evidencia técnica y trazabilidad. La calificación tributaria y la presentación ante la Agencia Tributaria deben revisarlas un asesor fiscal o profesional autorizado.
Open crypto tax audit docs
Checklist de expediente
Lo que un caso cripto/NFT necesita antes de ponerse delante de un asesor.
Un expediente útil no es una captura de Etherscan. Es una cadena de fuentes, cálculos, supuestos y huecos pendientes que otra persona puede revisar sin adivinar.
Audit readiness
0/8 checks
Casos de buen uso
Paneles vivos para explicar cuándo CORTEX tiene sentido.
No todo necesita un ledger. CORTEX brilla cuando la memoria de un agente, una auditoría o una investigación puede convertirse en responsabilidad real.
Auditoría blockchain y NFTs para revisión fiscal.
Reconstruye wallets, exchanges, swaps, ventas NFT, royalties, fees y gaps de custodia en un expediente técnico trazable para asesor fiscal.
InputWallets + exchange CSVs + NFT marketplaces
CORTEXClassify, hash, flag gaps, seal evidence
OutputAdvisor-ready PDF, CSV, manifest
Flagship use case
MOSKVbot: agentic memory with a verifiable passport.
The strongest CORTEX story is not another memory API. It is a bot that can remember, prove, research, coordinate agents, and hand off state without turning chat history into the source of truth.
Portable context with proof.
MOSKVbot can carry user preferences, project decisions, open tasks, research crystals, error history, and agent handoff state as scoped, auditable records.
preferences
decisions
research
handoffs
ledger proofs
Pull tenant-scoped context before the bot answers.
Separate facts, conjecture, provenance, and stale claims.
Route work across agents with consensus and bounded authority.
Export signed evidence instead of unverifiable transcript memory.
Regular memory bot
Remembers useful context, but trust depends on the provider and the transcript.
MOSKVbot on CORTEX
Remembers with provenance, scoped tenants, verifiable checkpoints, research trails, and agent handoffs that survive across tools.
Positioning
CORTEX vs. the rest.
CORTEX can coexist with hosted memory tools, vector stores, and agent frameworks. The difference is architectural: CORTEX treats memory as auditable state, not just recall infrastructure.
Architecture comparison.
CORTEX focuses on verifiable memory and evidence continuity. Vendor pricing and product scope can change, so teams should confirm current terms during due diligence.
| Capability | Mem0 | Zep | Letta | CORTEX |
|---|---|---|---|---|
| Cryptographic ledger | Not the core contract | Not the core contract | Not the core contract | SHA-256 hash chain |
| Merkle checkpoints | Not native | Not native | Not native | Batch integrity roots |
| Multi-agent consensus | Not primary surface | Not primary surface | Agent runtime focus | WBFT voting + reputation |
| Privacy posture | Cloud/service dependent | Cloud/service dependent | Self-host option | Local-first + ingress guards |
| AST / static gates | Outside memory layer | Outside memory layer | Outside memory layer | Static gates and ship checks |
| Local-first mode | Cloud-first | Cloud-first | Supported | SQLite local mode |
| MCP / operator surface | Integration-dependent | Integration-dependent | Agent framework surface | API + CLI + SDK + MCP |
| EU AI Act evidence | Customer assembled | Customer assembled | Customer assembled | Article 12 evidence support |
| Core license posture | Vendor plan | Vendor plan | Open-source available | Apache 2.0 core |
Adoption questions
Hard questions before adoption.
The landing should make the trust boundary explicit. These are the objections a security reviewer or technical founder will raise first.
Does CORTEX make us EU AI Act compliant?
No tool can make that claim alone. CORTEX supplies technical logging, traceability, and integrity evidence that can support a broader compliance program.
Do we need to send memory to CORTEX Cloud?
No. The core path is local-first. Hosted sync and managed evidence exports are optional paid surfaces, not a prerequisite for ledgered memory.
Can it sit next to existing vector stores?
Yes. Treat CORTEX as the admission and proof layer. Your vector database can still handle retrieval scale while CORTEX preserves provenance and trust state.
What happens when validation fails?
The write should fail closed. A rejected generative output stays rejected instead of becoming durable fact with a hidden warning in logs.
Vercel canonical + optional Cloudflare edge
One public deployment. Billing stays behind the API.
`cortexpersist.com` is the canonical Vercel surface for the landing, docs, API routes, and Stripe handoff. Cloudflare can remain as DNS/CDN or edge policy, but not as a second source of truth for the web.
Vercel web
Serves `saas_landing.html`, `saas.css`, static documentation, canonical routing, and production aliases for `cortexpersist.com`.
Vercel FastAPI
Exposes `api/index.py`, mounts `/v1/stripe/checkout` and `/v1/proof-marks`, and talks to Stripe/Redis with server-only environment variables.
Cloudflare optional
Can provide DNS, CDN, security headers, or edge policy. It should not require a second manual deploy of the product page.
Required env
STRIPE_SECRET_KEY=sk_live_...
STRIPE_PRICE_TABLE={"pro":"price_...","team":"price_..."}
UPSTASH_REDIS_REST_URL=https://...
UPSTASH_REDIS_REST_TOKEN=...
CORTEX_PUBLIC_ORIGIN=https://cortexpersist.com
Commercial canonical
`cortexpersist.com` should be the public product surface, pricing, and checkout entry.
Developer surface
`cortexpersist.dev` should carry docs, API references, SDK examples, and staging links.
Trust and community
`cortexpersist.org` hosts governance, research, security posture, responsible disclosure, and evidence material.
Open trust hub
Pricing
Start locally. Add managed help when needed.
The open-source core is free. Paid tiers are for teams that want hosted sync, implementation support, or managed compliance evidence.
Community
For local builders and early product experiments.
$0 forever
- Full CLI and Python API
- Single-tenant local mode
- Ledger verification
- Hybrid search
- Apache 2.0 source
Early access
Pro
For solo operators running serious agents.
$29 /month
- Encrypted multi-device sync
- Five tenant workspaces
- Ghost tracking and handoff
- Priority implementation support
- Hosted evidence exports
Team
For engineering teams with audit pressure.
$99 /month
- Unlimited tenant workspaces
- WBFT consensus workflows
- Compliance report templates
- Graph knowledge fusion
- Team onboarding session
Enterprise
For regulated deployments and dedicated infrastructure.
Custom
- On-premise architecture review
- Custom guard and tenant policy
- Audit evidence pack
- Security review support
- Dedicated implementation lane
Do not let agent memory become unverifiable production state.
Install locally, store one decision, and verify the ledger before you trust the next agent handoff.