Cybersecurity Intelligence Platform for Human-in-the-loop Enterprise Response
Open-source, NICE 2.0-aligned, multi-agent cybersecurity workforce framework.
CIPHER maps all 41 NICE Cybersecurity Workforce Framework (v2.1) work roles to configurable AI agent personas, deployable in three operational modes:
| Mode | Description |
|---|---|
| 🤖 Autonomous (T3) | Agent executes low-risk actions independently (enrichment, triage, ticketing) |
| 👥 Copiloted (T2) | Agent recommends, human approves before execution |
| 🧑 Human-Administered (T1) | Agent advises only; human executes high-impact actions |
Every agent has a cryptographic identity (SPIFFE/SPIRE), role-scoped memory, and all actions pass through a policy engine (OPA) with an immutable audit ledger.
┌─────────────────────────────────────────────────────────────────┐
│ CIPHER FRAMEWORK │
├──────────────────┬──────────────────┬───────────────────────────┤
│ ORCHESTRATION │ SECURITY PLANE │ MEMORY PLANE │
│ LangGraph │ SPIFFE/SPIRE │ Episodic (TimescaleDB) │
│ NATS JetStream │ OPA Policies │ Semantic (ChromaDB) │
│ Autonomy Gate │ Input Guard │ Procedural (Playbooks) │
│ Human Approval │ Behavioral Mon. │ Working (Redis) │
│ Trust Broker │ Audit Ledger │ Credentials (Infisical) │
├──────────────────┴──────────────────┴───────────────────────────┤
│ 41 NICE ROLE AGENT PERSONAS │
│ PD: Defensive · IR · Forensics · Threat · Vuln · Infra · Insider │
│ IN: Cybercrime · Evidence │ IO: SysAdmin · NetOps · +5 │
│ OG: CISO · Risk · Policy · +13 │ DD: Architect · SecDev · +7 │
├─────────────────────────────────────────────────────────────────┤
│ INTEGRATION PLANE (per-source API keys) │
│ SIEM: Splunk · Elastic · QRadar · Sentinel · Wazuh │
│ EDR: CrowdStrike · SentinelOne · Defender │
│ Vuln: Tenable · Qualys · Rapid7 · OpenVAS │
│ Intel: MISP · VirusTotal · OpenCTI · AlienVault │
└─────────────────────────────────────────────────────────────────┘
- Agent Identity: SPIFFE SVIDs (Ed25519, 1hr TTL, auto-rotated)
- Authorization: OPA deny-by-default RBAC/ABAC on all memory + tool access
- Memory Integrity: TimescaleDB hash chain (tamper-evident episodic log)
- Injection Defense: Input Guard — all external data in
DATA_CONTEXT, neverSYSTEM_INSTRUCTION - Audit: WORM append-only ledger, OCSF export, OpenTelemetry traces on every action
Threat vectors covered: prompt injection · agent impersonation · data scope violation · memory poisoning · orchestrator compromise · agentic insider threat · APT log manipulation · credential exfiltration · cross-tenant bleed · model fine-tune attack · dormant payload activation
cipher/
├── AGENTS.md # Jules/AI coding agent guidance
├── docs/
│ └── ARCHITECTURE.md # Full architecture specification
├── specs/ # Source of truth for all implementations
│ ├── personas/ # YAML spec per NICE work role agent
│ ├── memory/ # Memory plane specs
│ ├── security/ # Security component specs
│ ├── orchestration/ # Orchestration component specs
│ └── integrations/ # Connector specs
├── prompts/ # Agent system prompts + memory templates
│ └── {role_id}/
│ ├── system.md
│ ├── memory_inject.md
│ └── output_schema.json
├── src/ # Implementation (Jules-generated from specs)
├── tests/ # Test suite
├── helm/ # Kubernetes Helm chart
├── docker/ # Docker Compose (dev)
└── roadmap/
└── ROADMAP.md
| Layer | Technology |
|---|---|
| Orchestration | LangGraph + NATS JetStream |
| Agent Identity | SPIFFE/SPIRE |
| Policy Engine | Open Policy Agent (OPA) |
| Episodic Memory | TimescaleDB (PostgreSQL) |
| Semantic Memory | ChromaDB → Weaviate |
| Working Memory | Redis |
| Credential Store | Infisical |
| Observability | OpenTelemetry + Grafana |
| LLM Routing | LiteLLM (model-agnostic) |
| Auth (Operators) | Keycloak |
| Language | Python 3.12+ |
| Version | Focus | Target |
|---|---|---|
| v0.1 | Core + 4 PD agents | 8 weeks |
| v0.2 | Full security plane + 5 more agents | 14 weeks |
| v0.3 | All 41 NICE roles | 24 weeks |
| v0.4 | Enterprise hardening + Kubernetes | 32 weeks |
| v1.0 | Production release + community launch | 36 weeks |
See ROADMAP.md for details.
This repository is designed for implementation via Jules (Google's autonomous coding agent). See AGENTS.md for Jules guidance and the specs/ directory for implementation specs.
Phase 1 Jules issues: #1–#6
Agent Profiles: dcplatforms.github.io/cipher
All 41 NICE work role agent profiles — full KSAs, objectives, authorized tools by autonomy tier, memory access config, and live action counters.
MIT © 2026 CIPHER Contributors