FlowHelm

Your AI agent, your helm.

FlowHelm is a secure, multi-tenant AI agent orchestrator. Run personal AI agents for yourself or your whole team on a single server. Each user gets their own isolated, extensible agent — with built-in channels (Telegram, WhatsApp, Gmail, Calendar, voice) and a skills system for adding new capabilities, integrations, and workflows.

What Makes FlowHelm Different

Extensible agent platform. FlowHelm ships with built-in channels and skills, but it's designed to grow. Anyone can build custom skills, channel adapters, and tool integrations — install what you need, remove what you don't. Your agent does what you tell it to.

Podman rootless, not Docker. No root daemon. No shared process. Each user's containers run in separate UID namespaces — one user literally cannot see another's containers. Compliant-ready for PCI-DSS, SOC 2, and HIPAA.

Per-user credential proxy. API keys never enter agent containers. A dedicated proxy container holds credentials and injects them at the network layer. Prompt injection can't extract what the agent doesn't have.

Multi-tenant by design. flowhelm admin add-user provisions an isolated agent in 60 seconds. Resource limits enforced via cgroups v2. Not a bolt-on — built from day one.

Gmail Pub/Sub streaming pull. Real-time email notifications in 2-3 seconds. No exposed ports, no webhooks, no tunnels.

Voice commands. Send voice notes on Telegram or WhatsApp. FlowHelm transcribes and executes.

Quick Start

# On a fresh Ubuntu 24.04 VM
curl -fsSL https://flowhelm.ai/install.sh | bash
flowhelm admin init
flowhelm admin add-user yourname
ssh flowhelm-yourname@localhost
flowhelm setup

Requirements

Linux (production): Ubuntu 24.04 LTS, 4 GB RAM minimum, Podman 4.0+, Node.js 22+ macOS (development): Apple Silicon M1+, macOS Tahoe (26), 8 GB RAM minimum

Documentation

• Architecture — System overview, component responsibilities, data flow
• Installation — Install script, admin commands, user setup wizard
• Deployment — Always-on operation, crash recovery, backups
• Security Model — Isolation boundaries, threat model, compliance
• Podman Isolation — Rootless containers, UID namespaces, SELinux
• Credential Proxy — Per-user MITM proxy, credential injection
• Multi-Tenant — User lifecycle, resource limits, port allocation
• Database — PostgreSQL schema, message queue, crash recovery
• Memory — Cognitive memory system, embeddings, identity layer
• Sessions — Warm container lifecycle, PG backup/restore
• Agent Runtime — CLI and SDK runtimes, token optimization
• Channels — Telegram, WhatsApp, Gmail adapters
• Channel Container — Unified channel container, HTTP API
• Gmail Pipeline — Pub/Sub streaming, IMAP, email filtering
• Voice Pipeline — Service container, whisper.cpp transcription
• Auth Bridge — Headless OAuth, QR code auth flow
• Skills — SKILL.md format, registry, per-user store

License

FlowHelm uses a source-available license based on MIT.

Free for personal use — you may run FlowHelm with up to 2 managed users at no cost. The admin account used for installation and configuration does not count toward this limit.

Commercial license required for deployments with 3 or more managed users, or for offering FlowHelm as a hosted service. Contact [email protected] for commercial licensing.

Contributing

See CONTRIBUTING.md. Contributors must sign a CLA.