Automation-grade AI copilots for teams, creators, and businesses.
Helix is an intelligent digital companion built for users, creators, and businesses β combining automation, contextual memory, and a secure plugin system in one extensible platform.
βUnify your tools. Amplify your mind.β
Helix is a modular AI assistant platform designed to unify your tools, workflows, and intelligence into a single, secure, human-friendly experience.
Unlike traditional assistants or bots, Helix is built as a full-stack system that spans chat interfaces, automation, memory, observability, and secure plugin execution. It adapts to each user or organization with contextual awareness, intelligent orchestration, and powerful extensibility β across web, Discord, mobile, and even the OS level.
Helix isn't just reactive β it's proactive, modular, programmable, and privacy-conscious.
Helix serves as a natural-language assistant that can answer questions, manage workflows, run automations, and execute commands across your connected tools. It understands user intent, tracks goals, and provides reasoning with memory and persona context β not just static answers.
Powered by pgvector and long-term embedding memory, Helix tracks context per user, conversation, tool, and organization. It remembers what matters β conversations, facts, files, decisions β and lets you query or forget them at will. Think of it as a second brain with search and TTL.
Helix introduces a secure sandboxed runtime for custom tools, automations, and third-party extensions. Developers can publish skills using signed manifests, and users can run tools with explicit scopes like read:calendar or write:memory. This bridges the gap between assistant and operating system.
Helixβs UI is built to streamline how users interact with AI, memory, and tools. Features include a plugin drawer, toolcall inspection, persistent conversations, personas, and live updates. Whether via web, mobile, or Discord β the experience is consistent, assistive, and extensible.
Helix can route tasks dynamically across multiple model providers: OpenAI, Claude, Ollama (local), and fine-tuned internal models. This lets users choose between performance, privacy, and cost β all controlled by tier, policy, or inference health. Itβs an intelligent LLM load balancer and policy enforcer in one.
Users and orgs can view personalized dashboards with memory usage, plugin activity, skill latency, and AI model performance. These dashboards are powered by OTEL and Grafana, and support annotations, sharing, and custom metrics per tenant.
HelixBot brings the full assistant experience to Discord β combining Dyno moderation, TicketTool, Tupperbox, and AI agent capabilities. Users can create tickets, assign skills, run commands, or collaborate with Helix directly inside their communities.
Every part of Helix is accessible via its open API and TypeScript SDK (@helix/sdk). Developers can integrate memory, skill runs, automations, or chat experiences into their own tools, allowing Helix to act as a platform, not just a product.
Helix is engineered with enterprise-grade security from the start:
- Per-tenant secrets stored via Vault/KMS
- mTLS mesh for edge and plugin traffic
- Signed Audit Logs (Ed25519)
- RBAC/ABAC controls for every tool and API
- Air-gapped deployment support for compliance environments (SOC2, HIPAA, GDPR)
In short: Helix is not just another assistant β itβs a full-stack, programmable, privacy-first, user-centric AI operating system that runs anywhere: browser, Discord, mobile, or serverless edge.
| Type | Highlights |
|---|---|
| Memory | Short-term (Redis) + long-term (pgvector) with embedded provenance |
| Skills | Deno sandboxed plugin runtime with signed manifests and scoped permissions |
| Automations | Triggers β Filters β Actions (webhooks, schedules, events) |
| Security | RBAC/ABAC, per-tenant secrets, audit logs (Ed25519 signed) |
| Observability | OTEL tracing, Grafana dashboards, k6 load tests, Chaos mode |
| Chat UI | Multimodal assistant with persona overlays, memory viewer, skill drawer |
| API | Open REST API + TypeScript SDK (@helix/sdk) |
| Integrations | Discord, Stripe, GitHub, Google, Slack, Notion, and more (via marketplace) |
Helix is built on a modular, secure, multi-tenant architecture designed to scale across personal devices, cloud workloads, and enterprise environments β including fully air-gapped deployments.
- Framework: Next.js (App Router, Server Components)
- Styling: TailwindCSS + shadcn/ui for clean, component-driven UI
- Modes: Web app, Discord bot UI, upcoming mobile (React Native)
- UX Architecture: Chat interface + plugin drawer + memory viewer + dashboard panels
- Database: Supabase (Postgres + Realtime + Auth)
- ORM: MikroORM for schema-driven, typesafe Postgres modeling
- Cache & Queues: Redis (short-term memory, task queues, session context)
- Vector Memory: pgvector (HNSW/IVFFlat) for long-term memory & context embeddings
-
Router: Model control plane routes inference requests across providers based on cost, latency, org policy
-
Supported Models:
openai:gpt-4,openai:gpt-3.5-turboclaude-3-opus/claude-3-sonnetollama:llama3-70b,mistral, custom fine-tunes
-
Fallback Logic: Retry-on-failure, tier-aware switches, hybrid local/cloud override
- Primary Edge: Cloudflare Workers (webhooks, automations, inference)
- UI Delivery: Vercel Edge Functions + CDN
- Chaos Ready: Workers are instrumented for chaos testing (latency spikes, plugin sandbox failures)
- Secrets Backend: Vault, AWS KMS, or on-prem HSM per tenant/org
- Audit Trails: Append-only audit logs (Ed25519-signed)
- Plugin Isolation: Each skill/tool runs in a Deno sandbox with scope-limited memory and access
- RBAC/ABAC: Fully enforced at plugin, API, memory, and automation layers
-
Tracing: OpenTelemetry (OTEL)
-
Dashboards: Grafana Cloud, with panels for:
- Skill execution latency
- Plugin failures
- Memory recall precision
- Usage metrics per org/user/tier
-
Telemetry Stack:
Tempofor tracesLokifor logsMimirfor metricsPyroscopefor CPU profilesk6for load testingAlertmanagerfor incident ops
-
Deployment: GitOps-based (GitHub Actions β Vercel + Cloudflare deploys)
-
Multi-Tenant Support:
- Per-org data isolation
- Custom memory scopes (
org|user|tool|conversation)
-
Air-Gapped Mode:
- BYO Postgres, Redis, KMS, Workers
- No public SaaS dependencies
- CLI-based provisioning via
helixctl(WIP)
-
Plugin Runtime: Deno VM, signed manifest, permissions + review queue
-
Automation Engine: Trigger (event/webhook/schedule) β Filter β Action (tool/skill/memory)
-
SDKs:
- TypeScript:
@helix/sdk - Python (beta)
- REST API (OpenAPI Spec WIP)
- TypeScript:
-
Marketplace: Users and orgs can install 3rd-party tools securely
/apps
frontend # App Router UI with Tailwind + shadcn/ui
frontend-e2e # Cypress smoke + regression specs for the UI
services/
user # NestJS user/domain service (auth, accounts, org graph)
user-e2e # Jest e2e targets for the user service
/libs
config # Runtime config objects, secrets, telemetry helpers
db # MikroORM entities, repositories, and base models
hypertune # Hypertune feature-flag client + React bindings
ui # Shared component system, providers, and theme tokens
/Docs
Features.md # Product tiers and capabilities
DEPLOY.md # Deployment notes and cloud prerequisites
ThirdParty.md # Vendor/third-party integration referencesHelix supports a plugin manifest format for external tools.
pnpm plugin run ./plugins/my-plugin.tsSkills run inside a Deno sandbox with scoped permissions (read:calendar, write:memory, etc).
Use /docs/plugin-guide.md (coming soon) to build, sign, and test new tools.
| β / βοΈ / π§± | Milestone | Description |
|---|---|---|
| βοΈ | Core Memory System | Implemented contextual memory with Redis (short-term) and pgvector-backed long-term embeddings, organized by MemorySpace and MemoryShard. Includes time-bounded retention and persona-aware recall. |
| βοΈ | Hybrid Inference Router | Fully operational routing layer that dynamically switches between OpenAI, Claude, and Ollama/local models. Supports fallback logic, model policy enforcement, and latency- or tier-based switching. |
| βοΈ | Plugin Runtime MVP | Sandbox-ready Deno VM to run signed tools with manifest-defined permissions. Current focus: execution tracing, audit logs, scoped memory access. Used for skills like βschedule.postβ, βsummarize.pageβ, etc. |
| βοΈ | Web + Discord UI | Unified assistant interface across web (Next.js + shadcn/ui) and Discord (HelixBot) with support for chat, memory inspection, tool drawer, personas, and message context. |
| π§± | Chaos Engineering Toolkit | Controlled failure injection for memory, inference, automations, and plugin runtime. Includes OTEL-backed observability, rollback hooks, and pre-prod harness. Will support chaos_mode_enabled flag. |
| π§± | IoT Edge Agent + Smart Home Control | MQTT/Zigbee-capable Agent model with support for edge heartbeat, command dispatch, and state sync. Helix will control devices via text or scheduled automations. |
| π§± | Helix Linux Distro | Privacy-first Linux environment with Helix as the core AI assistant. Includes native CLI (helixctl), voice activation, secure local inference, and offline plugin support. |
| π§± | Mobile App (Android/iOS) | React Native-based assistant interface with chat, notifications, memory sync, and voice input. Will support offline caching, push messaging, and native intents. |
| π§± | Developer Marketplace | Plugin marketplace where creators can submit, review, and share skills, integrations, and workflows. Includes signed manifest submission, sandbox tests, and scoped install UI. |
Pull requests, issues, and contributions are welcome! Please follow the Contributor Guidelines and review the Code of Conduct.
Polyform Noncommercial License 1.0.0 Β© SinLess Games LLC
Helix is a source-available platform. Noncommercial use is permitted under the Polyform license.
Commercial use, resale, or sublicensing requires a separate license from helixaibot.com.
Made with β€οΈ by @Sinless777 and contributors
