Comparison

humux vs.
the personal AI ecosystem

The self-hosted agent space is growing fast. Here's how humux stacks up against the other major personal AI agents — honestly.

Hermes

Nous Research — 207k ★

Python agent with memory, scheduling, browser automation, subagents, MCP, and voice. Self-hosted, Docker or bare metal. Rich plugin ecosystem.

OpenClaw

OpenClaw — 381k ★

TypeScript agent with 25+ channels (WhatsApp, Telegram, Signal, iMessage, Discord, etc.). Cross-platform desktop. Largest ecosystem.

ZeroClaw

ZeroClaw Labs — 32k ★

Rust agent runtime. Single binary, 30+ channels, 20+ LLM providers, hardware I/O (GPIO/I2C/SPI), SOP engine, web dashboard. Fast & minimal.

NemoClaw

NVIDIA — 21k ★

Reference stack for running agents (OpenClaw, Hermes, LangChain) inside NVIDIA OpenShell sandboxes. Adds managed inference, network policy, lifecycle.

TrustClaw

Composio — 845 ★

Vercel-deployed agent with 1000+ Composio tools via OAuth. Sandboxed execution. Telegram + web UI. Security-first, zero local config.

humux

Merola Software & Services

Python agent with built-in email, calendar, contacts, Telegram, voice, vault, skills, subagents, and admin UI. Single Docker container.

Feature humux Hermes OpenClaw ZeroClaw NemoClaw TrustClaw
Self-hosted Sandbox Vercel
Single container deploy Vercel
Messaging channels 3 ~5 25+ 30+ Depends 1
Telegram (per-agent bots) Single bot Via agent
WhatsApp Via bridge
CLI channel
Email Via tool
Calendar (CalDAV) Via tool
Contacts (CardDAV) Via tool
Persistent memory Via agent
Voice (STT + TTS)
Markdown skills (no-code) Via agent
Subagents Via agent
Scheduled tasks / cron
Web search Via agent
Browser automation
Encrypted secrets vault
Granular permissions Sandbox OAuth
Admin web UI
LLM providers 6 ~15 ~20 ~20 Routed Multiple
Open source license MIT MIT MIT Apache 2.0 Apache 2.0 MIT
Language Python Python TypeScript Rust TypeScript TypeScript

Info gathered from public repos and docs at time of writing. Stars as of mid-2026. Products evolve — check each project for current state.

Where humux stands out

Integrated productivity

Email, calendar, and contacts built in — not via third-party tool connectors. Per-agent mailboxes with vault-secured credentials. The only agent with native CardDAV and CalDAV.

Deep Telegram

Per-agent bots in one group — each replies only when addressed, no loops. Voice messages, inline approvals, reactions, per-chat permissions. Plus WhatsApp via wacli.

Markdown skills

Teach the agent by writing markdown files — no code, no plugin registry. Add CLI tools by describing their command syntax. Skills ship as plain files in skills/.

Security model

Encrypted two-tier vault (infra keys + agent secrets). Credentials never enter LLM context. Glob-pattern permissions with ALWAYS/ASK/NEVER. Telegram inline approvals. No telemetry.

Honest trade-offs

Every tool has its strengths. Here's where each project shines — and where humux makes different choices.

🦞

OpenClaw

github.com/openclaw/openclaw — 381k ★

The gorilla in the room. 25+ messaging channels (WhatsApp, Telegram, Signal, iMessage, Discord, etc.), cross-platform desktop apps (macOS, Linux, Windows), and the largest ecosystem of any personal AI agent. Sponsored by OpenAI, GitHub, NVIDIA, and Vercel.

humux trades channel breadth for depth: fewer channels but per-agent Telegram bots, native email/calendar/contacts, markdown skills, and an encrypted vault. OpenClaw is the better choice if you need 20+ messaging platforms.

🧠

Hermes

NousResearch/hermes-agent — 207k ★

The agent that grows with you. Rich plugin/skills ecosystem, browser automation, subagents, MCP server support, voice, and self-evolution (optimizes its own skills via DSPy). Strong Nous Research backing. Active community.

humux focuses on different integrations: built-in email/calendar/contacts vs Hermes's broader plugin ecosystem. Hermes has more LLM providers and a larger community. humux has a stronger out-of-the-box productivity feature set.

🦀

ZeroClaw

zeroclaw-labs/zeroclaw — 32k ★

A Rust binary — tiny, fast, single-file deploy. 30+ channels, 20+ LLM providers, hardware I/O (GPIO, I2C, SPI on Raspberry Pi), SOP engine, web dashboard, and OS-level sandboxing (Landlock, Bubblewrap, Docker).

humux trades raw performance for feature depth: Python gives us richer libraries for email/calendar/voice. ZeroClaw is the right choice for edge/embedded or if you need the smallest possible footprint.

🔬

NemoClaw (NVIDIA)

NVIDIA/NemoClaw — 21k ★

Not an agent — a reference stack for running agents inside NVIDIA OpenShell sandboxes. Adds managed inference, network policy, sandbox lifecycle, and credential handling. Runs OpenClaw, Hermes, or LangChain agents in hardened environments.

Different layer of the stack: NemoClaw is infrastructure for securing agent execution. humux is a standalone agent you can run anywhere. For enterprise NVIDIA GPU environments, NemoClaw + an agent makes sense. For a single Docker container on any host, humux ships complete.

🔒

TrustClaw (Composio)

ComposioHQ/trustclaw — 845 ★

Security-first agent with 1000+ tools via OAuth (Gmail, GitHub, Slack, Notion, etc.) and sandboxed execution on Composio's cloud. One-click Vercel deploy. No local config, no API keys in config files. Cron scheduling, Postgres-powered memory, Telegram + web UI.

Different threat model: TrustClaw's sandboxed remote execution is safer for untrusted tool use. humux runs everything locally — faster and offline-capable, but requires you to manage secrets and permissions. TrustClaw's Vercel deploy isn't truly self-hosted; humux is fully air-gappable.

humux

github.com/mattmezza/humux

The only agent with built-in email, calendar, and contacts — not via tool plugins. Deep Telegram with per-agent bots in shared groups. Markdown skills: teach by writing files. Encrypted vault with agent/infra separation. Full admin UI. One container, docker compose up.

Best fit for: developers and privacy-conscious users who want a complete personal AI that handles messaging, email, calendar, and automation out of the box. humux is the most self-contained personal productivity agent in the ecosystem.

Run your own personal AI

One container. All integrations built in. Your data stays yours.