Overview

HELIX is an open-source AI engineering operating system packaged as an npm CLI. It installs a structured team of 65 specialist AI agents into any repository with a complete runtime for planning, execution, handoffs, and enterprise governance.

What Makes HELIX Different

| Capability | HELIX | Basic AI Assistants | |---|---|---| | Agent team | 65 specialists across 7 domains | 1 generalist | | Execution | Python (deterministic) → NPX (LLM API) → Manual | Manual only | | Process | Requirements → Architecture → Review → Code | Jump to code | | Security | Input guard, PII detection, RBAC, sandbox | None | | Audit | Immutable SHA-256 hash-chained audit trail | None | | Deployment | Multi-tenant, REST API, Grafana dashboards | Local only | | Migration | 7 dedicated migration agents (DB, Cloud, OnPrem, etc.) | None |

Agent Teams

• Core (8): Orchestrator, Memory, Governance, Help, Agent Forge, Debate Mode, Prompt Engineer, Repo Doctor
• Architecture (6): Project Architect, Solution Architect, Domain Modeler, System Integration, Business Analyst, Design System
• Engineering (5): Frontend, Backend, Engineering Review, UI/UX, Accessibility
• Security & Compliance (5): Security, Privacy, Threat Modeler, Identity & Access, API Governance
• QA & Reliability (4): QA, Test Automation, Integration Test, SRE, Chaos Resilience
• Infrastructure (7): DevOps, Cloud, Platform, Observability, FinOps, IaC, Kubernetes
• Data & AI (8): Data Engineer, Data Governance, Data Quality, ML Engineer, AI Architect, Analytics, Database, DB Migration
• Product & Knowledge (8): Product Strategy, Content Strategist, Technical Writer, Release Manager, Open Source, Legal, Example Builder, Knowledge
• Enterprise (7): Customer Success, Procurement, Workflow, Performance, Mobile, Accessibility, Database Migration
• Migration (7): Orchestrator, OnPrem→Cloud, Database, Cloud→Cloud, Security, Validation, Rollback

Three-Tier Execution Engine

1. PYTHON (instant, zero LLM cost)
   helix run <session> --engine python
   → Deterministic file scanning, pattern matching, anti-pattern detection
   → 15+ agents have Python fast-path rules

2. NPX AUTO (LLM API, ~$0.01-0.05/agent)
   helix plan "build X" --auto
   → Full auto: plan + execute all agents via API
   → Circuit breaker, retry, model fallback, budget guard

3. MANUAL (copy/paste to any LLM tool)
   helix run <session>
   → Formatted prompt for Cursor / Claude Code / Copilot / Codex
   → Human-in-the-loop with full control

Slash Commands — 65 Agents at Your Fingertips

Type /helix-security, /helix-database, /helix-migration-orchestrator in your AI tool and the agent's full SKILL.md loads as context:

• Claude Code: .claude/commands/helix-security.md
• Cursor: .cursor/commands/helix-security.md
• GitHub Copilot: .github/prompts/helix-security.md

Enterprise Runtime

• REST API: helix serve — GET/POST sessions, agents, tenants, health
• RBAC: viewer / reviewer / executor / admin with 25 granular actions
• Multi-tenancy: ai/tenants/<id>/ isolation for SaaS deployments
• Audit: Immutable SHA-256 hash-chained JSONL audit trail
• Observability: OpenTelemetry metrics, structured JSON logging, Grafana dashboard
• Compliance: PII detection, SOC2/GDPR ready, retention policies, right-to-delete

Full documentation: https://helix.datainteg.io/

Install

Use HELIX directly with npx:

npx @datainteg/helix

Install all agents non-interactively:

npx @datainteg/helix install --preset full --yes

Install into a specific project:

npx @datainteg/helix install --preset startup-saas --target C:\work\factory-maintenance-platform --yes

Validate the installation:

helix doctor
helix list --presets

What Gets Installed

HELIX treats .helix/skills as the source of truth and generates tool-specific targets during install.

.helix/skills/      source agent skills
.claude/skills/     generated Claude skill target
.claude/commands/   generated Claude slash commands: /helix, /helix-help
.agents/skills/     generated generic agent target
.agents/commands/   generated generic command prompts
HELIX/               runtime planner, schemas, prompts, providers, config
ai/memory/           shared project memory
ai/sessions/         runtime sessions
ai/handoffs/         agent-to-agent handoff packets
ai/reports/          delivery reports
ai/reviews/          specialist review artifacts
ai/usage/            token and cost logs

Generated .claude and .agents folders are not tracked in this package repository. They are produced in target projects by helix install.

Quick Start

Manual mode needs no provider API key.

helix plan "helix"
helix plan "build SaaS app with auth dashboard payments"
helix run <session-id>
helix run <session-id> --target cursor
helix run <session-id> --target claude
helix run <session-id> --target codex

For tools that support markdown slash commands:

/helix
/helix-help

/helix starts requirement intake and staged planning. /helix-help explains workflow direction, recommended agents, prompt examples, output locations, and next commands.

End-To-End Workflow

flowchart TD
  A[User goal or /helix] --> B[Requirement understanding]
  B --> C[helix-help guidance]
  C --> D[Agent selection and stage plan]
  D --> E[Runtime session JSON]
  E --> F[Specialist agent execution]
  F --> G[Handoff packets]
  G --> H[Reviews, memory, reports]
  H --> I[Validation and release decision]
  G -->|scope change| B
  G -->|conflict| J[helix-orchestrator]
  J --> D

Industrial Example

Scenario: an industrial manufacturing company wants a predictive maintenance platform for factory equipment. The platform needs sensor ingestion, anomaly detection, technician dashboards, alerting, audit history, secure access, cloud deployment, and operational monitoring.

1. Start with guidance

In Claude Code or a compatible tool:

/helix

Or through the CLI:

helix plan "helix"

Expected guidance:

• Clarify factory users, equipment types, sensor sources, uptime goals, and deployment limits.
• Recommend helix, helix-help, helix-business-analyst, helix-project-architect, helix-data-engineer, helix-ai-architect, helix-security, helix-sre, and deployment specialists.
• Ask missing questions before implementation begins.

2. Create the real execution plan

helix plan "Build an industrial predictive maintenance platform for factory machines with IoT sensor ingestion, anomaly detection, technician dashboard, alert notifications, role-based access, audit history, cloud deployment, and SLO monitoring"

HELIX creates:

• ai/sessions/<session-id>.json
• ai/handoffs/*.json
• ai/reports/final-delivery-<session-id>.md

3. Run agents manually in your AI tool

helix run <session-id> --target codex

Paste the generated prompt into Codex, Claude Code, Cursor, or another AI coding tool. After the current agent completes:

helix run <session-id> --complete --summary "requirements and missing questions captured"
helix run <session-id> --target codex

4. Expected stage flow

flowchart LR
  R[Requirement Intake<br/>helix, helix-help, business analyst]
  P[Preflight<br/>repo doctor]
  A[Architecture<br/>project, solution, domain]
  D[Data and AI<br/>data engineer, governance, quality, AI architect]
  S[Security<br/>security, threat model, identity, privacy]
  Q[Reliability<br/>QA, tests, database, SRE]
  C[Cloud Delivery<br/>DevOps, cloud, IaC, observability]
  X[Release<br/>docs, release, customer success]
  R --> P --> A --> D --> S --> Q --> C --> X

5. Run auto mode when ready

Auto mode uses your own provider key:

helix config set mode auto
helix config set provider openai
helix config set model gpt-4.1-mini
set OPENAI_API_KEY=your-key
helix run <session-id> --auto --continue --max-steps 3

Use a low --max-steps value first so the team can review early agent outputs before continuing.

Agent Presets

helix install --preset core
helix install --preset frontend
helix install --preset backend
helix install --preset infra
helix install --preset data-ai
helix install --preset startup-saas
helix install --preset open-source
helix install --preset full

Preset intent:

• core: intake, help, orchestration, memory, governance, validation, prompt quality, debate, and agent creation.
• frontend: UI, UX, accessibility, performance, frontend implementation, and browser-level quality.
• backend: APIs, services, contracts, identity, database, migrations, and integration tests.
• infra: cloud, IaC, Kubernetes, platform, observability, SRE, resilience, workflow, and cost.
• data-ai: data engineering, governance, quality, ML, AI systems, analytics, and privacy.
• startup-saas: practical SaaS delivery team for product, engineering, security, deployability, docs, and adoption.
• open-source: package readiness, docs, licensing, examples, release, support, and community trust.
• full: all built-in HELIX agents.

Core Commands

helix init
helix install --preset full
helix list
helix list --presets
helix plan "build an internal API platform"
helix run <session-id>
helix config get
helix config set provider openai
helix usage
helix export cursor
helix export claude
helix export codex
helix doctor
helix validate
helix status
helix --version

Provider Configuration

Supported provider IDs:

• openai
• anthropic
• gemini
• deepseek
• groq
• ollama
• openai-compatible

Keys are never stored in .helix/config.json. Use environment variables:

OPENAI_API_KEY
ANTHROPIC_API_KEY
GOOGLE_API_KEY
DEEPSEEK_API_KEY
GROQ_API_KEY
OLLAMA_BASE_URL
OPENAI_COMPATIBLE_API_KEY
OPENAI_COMPATIBLE_BASE_URL

Repository Layout

bin/                  CLI entrypoint
lib/                  install, registry, doctor, slash command generation
HELIX/registry/       58 agents and presets
HELIX/runtime/        planner, sessions, handoffs, prompts, LLM providers, usage
.helix/skills/        packaged agent source
ai/                   memory and runtime folder templates
docs/                 package documentation and logo assets
examples/             example workflows
templates/            project templates

Development

npm install
npm run check
npm run smoke
npm run validate
npm run test
npm run pack:dry-run

Release Gate

Before publishing:

npm run test
npm run pack:dry-run
npm publish --access public

Links

• Docs website: https://helix.datainteg.io/
• Issues: https://github.com/datainteg/helix/issues
• Support: [email protected]