Skip to content
@arkelythex

ARKELYTHEX

ARKELYTHEX — Fiscal intelligence & autonomous systems. Drenyra command center, evidence-first compliance, and AI-native infrastructure for the agentic era.
  • Peru

ARKELYTHEX

Mission-grade intelligence systems for organizations operating in Latin America.

Drenyra AI — alpha v0.5.0 Drenyra Pi — pre-alpha

ARKELYTHEX builds the Drenyra Accounting Ecosystem: a verifiable financial operating system where fiscal correctness, tenant isolation, and auditability are product safety requirements. Web, CLI, agents, and memory share one discipline — receipts prove execution, contracts freeze behavior, and a human accountant remains the final authority.


DRENYRA ACCOUNTING ECOSYSTEM

The core is six repositories with strict, explicit boundaries:

Repo Role Type Status
drenyra-command-center Accounting Command Center — product UI, tenants, documents, close, reconciliation, approvals, SUNAT Product / Command Center In development
drenyra-ai Verifiable Accounting Agent Ecosystem — protocol, runtime, receipts, ledger, missions, candidates, review Runtime and protocol Alpha (v0.5.0)
drenyra-pi Pi-native Accounting Operations Harness — operator experience, pinned verified runtime Pi-native harness Pre-alpha
drenyra-engram Institutional Accounting Memory — scope-first memory, lifecycle, provenance Institutional memory Alpha (v0.2.1)
drenyra-skills Versioned accounting, tax, and operational knowledge — content layer (content ≠ runtime) Knowledge In development
drenyra-guardian-angel Independent adversarial verification — refutation, dual review, evidence checks Verification In development

Drenyra Accounting Ecosystem Drenyra Accounting Ecosystem

Open the interactive version for pan, zoom, focus, and relationship tracing.

Dependency direction

  • drenyra-command-center consumes released, versioned drenyra-ai; uses drenyra-engram for context.
  • drenyra-pi bundles foundation/operator skills and consumes a pinned, package-local, checksum-verified drenyra-ai (never PATH); uses drenyra-engram.
  • drenyra-ai never depends on the Command Center or Pi, and never knows Drenyra Pi exists.
  • drenyra-engram is independent — it informs, it never authorizes.
  • drenyra-skills owns specialized, versioned fiscal and jurisdictional knowledge as content, not runtime code; drenyra-ai validates and pins it for consumers.
  • drenyra-guardian-angel verifies against the frozen drenyra-ai contracts — never the author of what it reviews.

Skills packaging boundary

Drenyra takes inspiration from Gentle-AI's packaging model: the Pi package ships the stable skills needed to operate the harness, while specialized knowledge remains an independently versioned catalog.

  • Bundled in drenyra-pi — foundation and operator skills for scope, evidence, chains, safety, and the Pi workflow.
  • Maintained in drenyra-skills — Peru and future jurisdictional, regulatory, and practice-specific knowledge, with sources, validity, and version history.
  • Validated by drenyra-ai — runtime checksums, signatures, pinning, and mission authority; neither skills repository authorizes fiscal actions.

This keeps the package convenient without making the Pi harness the owner of the complete fiscal-knowledge catalog.

Authority model

A fiscal close is proposed by the runtime, authorized by a human, committed to PostgreSQL, and proved by a signed receipt that an independent verifier re-reads.

Fiscal close: candidate to signed receipt Fiscal close: candidate to signed receipt

Open the interactive version.

Drenyra accounting database → transactional truth (PostgreSQL, tenant-isolated)
Drenyra Engram             → institutional memory (guides, never authorizes)
Drenyra AI receipts+ledger → execution proof (Ed25519-signed, append-only)
Drenyra Guardian Angel     → independent adversarial verification (never the author)
Drenyra Pi                 → operator experience
Human accountant           → final authority
  • The runtime proposes, never decides: Drenyra AI reads tenant-scoped records and builds a candidate close, and nothing reaches PostgreSQL until a human authorizes it.
  • What the receipt proves: every committed action appends an Ed25519-signed receipt, and money is integer math — the committed cents are the receipt's cents.
  • Verification is independent: Guardian Angel reads frozen receipt bytes, not live state, and a challenged receipt is a finding, never a silent overwrite.

Evidence pipeline

Every committed action carries its own evidence from the moment it is extracted.

Evidence pipeline Evidence pipeline

Open the interactive version.

  • Two records, one truth: PostgreSQL holds current transactional truth, isolated per tenant, and the append-only ledger holds immutable proof of how it got there.
  • Evidence is produced, not written later: a candidate carries its own evidence from extraction onward, and the receipt is signed at commit time, never reconstructed afterwards.
  • Verification has a separate owner: Guardian Angel reads frozen bytes and can only advise or challenge; disagreement is a recorded finding that a human resolves.
  • Memory informs, it never authorizes: Drenyra Engram stores durable institutional history and provenance, and can never approve, commit or override a fiscal action.

Principles

  • Mission-grade — idempotent operations, total traceability, graceful degradation.
  • Evidence-native — every action produces verifiable evidence automatically (RDA receipts, audit ledger).
  • Human command — AI assists and prepares; humans authorize (approval tiers: single and dual); systems execute.
  • Deterministic fiscal truth — fiscal paths are test-covered; money is integer math (BigInt cents), never floats.
  • Scope-first — company/RUC/period isolation is structural in queries, memory, and search.
  • Contracts as public surface — versioned, transport-agnostic, with canonical vectors and migration paths.

THE ARKELYTHEX FIVE-LAYER MODEL

Every ARKELYTHEX product is built on the same five layers. Each layer abstracts one concern and can operate independently; a higher layer is enabled by the one beneath it, never by skipping one.

ARKELYTHEX five-layer model ARKELYTHEX five-layer model

Open the interactive version.

  • Layer 1 · Fabric — shared infrastructure: identity, tenant isolation, permissions, events, documents, audit, sync.
  • Layer 2 · Ontology — the live model of the operation: objects, relations, states, actions, policies, provenance.
  • Layer 3 · Intelligence — agents, reasoning, document extraction, classification, planning, semantic search.
  • Layer 4 · Command — missions, alerts, approvals, incidents, workflow, human oversight and override.
  • Layer 5 · Edge — devices, sensors, cameras, drone inspection, on-device inference, offline operation.

Drenyra and Elvyra operate across layers 1–4 today. Layer 5 lands next, and only after layers 1–4 are mature.


ARKELYTHEX'S OTHER SYSTEMS

Repo Role Status
.github Organization profile and community health Active
admin Internal tools and administrative infrastructure Maintenance / review
elect-validate Electoral act validation suite for Peru's ONPE (Go, CLI, web) Maintenance / review

Future verticals (industrial operations, public sector, edge) are planned but have no public repositories yet — they will land as clearly-identified repos, never mixed into drenyra-command-center.


Current priorities — Q3 2026

  1. Drenyra AI ecosystem maturity — Phases 2c (hybrid orchestration) + 3 are current; expose agents/ as a package subpath, keep releasing with provenance (SHA-256 manifest, SBOM, signed tag) and drive toward v1.0.
  2. Drenyra Engram engine — deliver Evidence Objects (v0.7.0), Evidence Lifecycle (v0.8.0), and Review Workspace (v0.9.0); freeze strategy on neutral contracts + PostgreSQL authority.
  3. Drenyra Pi harness — land runtime pin 0.3.0 (fiscal harness extraction complete), /drenyra:status + /drenyra:doctor end-to-end.
  4. Drenyra scope discipline — no new non-accounting verticals enter the product repo; progressive retirement of foreign folders.

The six Drenyra repositories are public and define the current production architecture. Other production repositories may remain private while public contracts, specifications, threat models, and reference slices are released progressively. Everything else on this page is auxiliary or in transition.

All architecture diagrams on this page are generated with Archify from versioned specifications in docs/diagrams.

Popular repositories Loading

  1. admin admin Public

    Administración — Internal tools and administrative infrastructure for ARKELYTHEX operations.

    TypeScript

  2. elect-validate elect-validate Public

    Electoral act validation suite for Peru's ONPE with Go concurrency, CLI and web UI.

    JavaScript

  3. .github .github Public

    ARKELYTHEX organization profile and community health

  4. drenyra-command-center drenyra-command-center Public

    Drenyra — command center for ARKELYTHEX: fiscal intelligence dashboard (React 19), API orchestration (Elysia), Go TUI (Bubbletea), agent orchestration (Mastra), and evidence-first AI infrastructure.

    TypeScript

  5. drenyra-engram drenyra-engram Public

    Institutional Accounting Memory — scope-first search, relations, lifecycle, provenance and institutional knowledge for Drenyra. Pre-alpha.

    Go

  6. drenyra-ai drenyra-ai Public

    Verifiable Accounting Agent Ecosystem — Receipt-Driven Accounting, candidates, proportional review, gates, ledger and recovery for Drenyra. Pre-alpha.

    TypeScript

Repositories

Showing 9 of 9 repositories

People

This organization has no public members. You must be a member to see who’s a part of this organization.

Top languages

Loading…

Most used topics

Loading…