← Back to Home

Zilli

Enterprise AI Agent self-development, privacy governance, and RL training framework

Overview

Zilli is a next-generation Agent tool engineering framework for AI self-development and enterprise privacy governance. Its core philosophy is "AI writes AI", "evaluation as development", "from the environment", and "from Agent to RL". The system combines a five-phase RL training pipeline — forming a self-evolving closed loop — with enterprise-grade data classification (5 levels), hybrid local-cloud execution, privacy gatekeeping, and compliance reporting (GDPR/HIPAA/SOC2).

Five-Phase Architecture

Phase 1: Definition

Define auto-verifiable task sets and type-safe Agent API contracts based on Pydantic.

Phase 2: Data

Build simulated sandbox environments, generate trajectory data, and establish a layered experience replay pool.

Phase 3: Infra

Deploy heterogeneous computing (SGLang + Megatron-LM) with adaptive length control and async rollout scheduling.

Phase 4: Model

CISPO algorithm for stable multi-turn Agent training, with RLVR reward shaping and GRPO baseline support.

Phase 5: Evolve

Offline evolution engine (DSPy + GEPA) optimizes Skills, with continuous learning from production data.

Enterprise Privacy

Zilli's privacy module provides end-to-end data governance for sensitive enterprise workloads. Five data classification levels (PUBLIC, INTERNAL, CONFIDENTIAL, RESTRICTED, REGULATED) drive automated policy enforcement across the entire Agent lifecycle.

DataClassifier

5-level data classification with automatic PII/PHI detection. Any data containing personal identifiers is automatically elevated to at least CONFIDENTIAL level.

PolicyStore

Tenant-scoped privacy policies persisted in JSONL format. Supports allow/deny rules per data classification level, integration points, and retention configurations.

ReIDAssessor

Re-identification risk assessment for de-identified datasets. Evaluates singleton, k-anonymity, and linkage risks, flagging any residual re-identification probability above 5%.

ConsentManager

Granular consent tracking at the data-point level. Supports withdraw, expiry, and purpose-based consent queries. Full audit trail of consent lifecycle events.

PrivacyEngine

Orchestration engine that coordinates classifier, policies, consent, and reID assessment into a unified decision pipeline. Exposes a single evaluate() entry point for the entire privacy stack.

Hybrid Execution

Zilli's hybrid execution layer bridges local and cloud model deployment with privacy-preserving routing. Sensitive data stays local; sanitized payloads may use cloud models when appropriate.

PrivacyGatekeeper

Makes local/cloud/deny decisions based on data classification and tenant policy. RESTRICTED and REGULATED data is forced to local execution; CONFIDENTIAL data may route to cloud after sanitization.

Sanitizer

Redacts PII/PHI from data payloads before cloud dispatch. Uses configured redaction strategies per data type: mask, hash, replace, or remove.

HybridExecutor

Executes model inference with automatic local-first routing. Falls back to cloud only when the local model is unavailable and the data has been successfully sanitized.

Compliance & Governance

Zilli generates compliance reports from existing audit trails (JSONL format) for regulatory frameworks out of the box. No additional storage infrastructure required.

GDPR

Right to erasure audit, consent records, data processing activity logs, and cross-border transfer documentation.

HIPAA

PHI access logs, minimum necessary use verification, and breach notification readiness assessment.

SOC2

Security monitoring evidence, access control logs, change management records, and risk assessment summaries.

Maintained by Ethercoin

Zilli is now maintained by Ethercoin, a decentralized AI compute network that connects compute requestors, workers, and attestors through ZK+TEE+PoRW three-dimensional verification and on-chain settlement. With 12,847+ active nodes and 356 PFLOPS of total compute capacity, Ethercoin provides the infrastructure backbone for Zilli's AI agent training and evolution pipeline.

Ethercoin Network

Decentralized AI compute platform.

Provides distributed GPU/CPU resources for Zilli's RL training, model inference, and Skill evolution — all verified through ZK proofs and TEE enclaves.

Zilli

AI Agent training infrastructure layer.

Built on Ethercoin's distributed compute network, Zilli trains self-evolving agents with enterprise-grade privacy governance, compliance reporting, and hybrid local-cloud execution.

Zilli-trained models also deploy to IClawOS — an AI-native Linux distribution — with production interaction data flowing back to Zilli for continuous improvement. Together, Ethercoin, Zilli, and IClawOS form a complete AI ecosystem: decentralized compute → agent training → edge deployment.

Tech Stack

Language

  • Python 3.11+

Core Libraries

  • pydantic
  • numpy
  • dspy-ai

RL Algorithms

  • CISPO
  • GRPO

Privacy

  • DataClassifier
  • PolicyStore
  • ReIDAssessor
  • ConsentManager
  • PrivacyEngine

Hybrid Execution

  • PrivacyGatekeeper
  • Sanitizer
  • HybridExecutor

Compliance

  • GDPR Reporting
  • HIPAA Reporting
  • SOC2 Reporting

Deployment

  • Local
  • Cloud
  • Hybrid
  • Docker

Repository