ZeroClaw: Autonomous AI Personal Assistant Infrastructure for Any Platform

Hermes Agent: Self-Improving AI Agent • Self-Hosted LLM 2026: Ollama vs vLLM vs LocalAI

What Is ZeroClaw? #

ZeroClaw describes itself as “fast, small, and fully autonomous AI personal assistant infrastructure, any OS, any platform — deploy anywhere, swap anything.” It isn’t a single chatbot app; it’s the plumbing for building one: a self-hosted gateway that wires an LLM of your choice into every channel you actually communicate through, then gives it a controlled way to act on your behalf.

🔗 GitHub: https://github.com/zeroclaw-labs/zeroclaw 🌐 Homepage: https://www.zeroclawlabs.ai/

Built in Rust and dual-licensed MIT / Apache-2.0, ZeroClaw crossed 32,000+ GitHub stars, with 4,600+ commits, 400+ open issues, and 280+ pull requests — evidence of an actively maintained project rather than a weekend demo. The most recent commit landed August 1, 2026, the day before this article.

The Core Idea: One Assistant, Every Channel #

Most “AI assistant” projects pick one surface — a Discord bot, a Telegram bot, a web widget — and stop there. ZeroClaw instead treats the channel as a pluggable component:

• 💬 30+ communication channels: Discord, Telegram, Matrix, email, voice, webhooks, CLI, and more
• 🧠 ~20 LLM providers: Anthropic, OpenAI, Ollama, and any OpenAI-compatible endpoint — swap models without rewriting your automations
• 🛡️ Supervised autonomy: sandboxing, tool “receipts,” and human-approval gates for higher-risk actions
• 🔩 Hardware access: GPIO, I2C, SPI, and USB support for embedded deployments
• 🌐 Web dashboard: an HTTP/WebSocket gateway for chat, memory browsing, config editing, cron management, and tool inspection
• ⚙️ SOP engine: Standard Operating Procedures with event triggers and approval workflows for repeatable, auditable automations

The pitch is that you configure the assistant’s brain and its guardrails once, and it can then answer you on whichever channel is convenient — Telegram from your phone, a webhook from a CI pipeline, or a CLI session on a server.

Architecture #

ZeroClaw’s architecture separates three concerns that most single-purpose bots bundle together:

Because the hardware layer (GPIO/I2C/SPI/USB) sits underneath the same tool-execution model, a ZeroClaw assistant running on something like a Raspberry Pi can, in principle, flip a relay or read a sensor through the same audited tool-call path it uses to send a Telegram message — not a separate, unaudited script.

Installation #

Quick install (Linux/macOS) #

curl -fsSL https://raw.githubusercontent.com/zeroclaw-labs/zeroclaw/master/install.sh | bash

From source #

git clone https://github.com/zeroclaw-labs/zeroclaw.git
cd zeroclaw
./install.sh

ZeroClaw also documents explicit support for Windows, FreeBSD, NixOS, and Docker, with an interactive quickstart wizard that walks through channel and provider configuration on first run.

Supervised Autonomy, Not Unrestricted Autonomy #

The project is explicit that “fully autonomous” doesn’t mean “unsupervised.” Three mechanisms back that up:

1. Sandboxing — tool execution can be constrained rather than running with full host access
2. Tool receipts — a record of what a tool call actually did, so autonomous actions stay auditable after the fact
3. SOP approval workflows — the Standard Operating Procedures engine can require a human sign-off before a triggered action runs, useful for anything higher-stakes than “reply to a message”

For a framework that’s willing to touch GPIO pins and USB devices, this audit trail matters more than it would for a pure chatbot.

Use Cases #

1. A Single Assistant Across Every App You Use #

Configure one LLM-backed assistant once, then reach it from Discord at your desk, Telegram on your phone, or email when you’re offline — same memory, same tools, same guardrails.

2. Home Automation With an Audit Trail #

Run ZeroClaw on a Raspberry Pi or similar SBC wired to GPIO-controlled hardware, and let the SOP engine gate physical actions (like unlocking something or toggling power) behind explicit approval.

3. Ops Bot for a Small Team #

Wire ZeroClaw into Discord/Matrix plus a webhook channel from CI, and let it triage alerts, answer runbook questions, and escalate through an approval workflow instead of paging a human for everything.

4. Provider-Agnostic Experimentation #

Because the model layer is swappable across ~20 providers, teams can A/B a task between, say, Ollama running locally and a hosted Anthropic model without touching channel or tool configuration.

ZeroClaw vs. a Single-Channel Bot Framework #

Related Repositories #

Related Articles #

• Hermes Agent: Self-Improving AI Agent That Evolves With You — another self-hosted agent framework, focused on continuous self-improvement rather than multi-channel deployment
• Self-Hosted LLM 2026: Ollama vs vLLM vs LocalAI — pairs well with ZeroClaw’s Ollama support for a fully local assistant stack
• herdr: The Terminal Multiplexer Built for Running Many AI Agents at Once — for orchestrating multiple coding agents rather than one multi-channel assistant

Conclusion #

ZeroClaw targets a gap most AI-assistant projects leave open: running one autonomous assistant consistently across every channel you actually use, backed by a swappable LLM layer and an audit trail for the actions it takes — down to hardware I/O on embedded devices. At 32,000+ stars, 4,600+ commits, and a commit history that’s still active day-to-day, it reads as production-oriented infrastructure rather than a proof of concept.

Best for: Developers and self-hosters who want one assistant reachable from Discord/Telegram/email/CLI alike, with real guardrails around what it’s allowed to do autonomously.

GitHub: https://github.com/zeroclaw-labs/zeroclaw

Recommended Infrastructure for Self-Hosting #

ZeroClaw is designed to run continuously as a gateway process — a small always-on VPS is a natural fit:

• DigitalOcean — $200 free credit for 60 days across 14+ global regions; enough to run a ZeroClaw gateway plus Ollama for weeks while you evaluate it.
• HTStack — Hong Kong VPS with low-latency access from mainland China. This is the same IDC that hosts dibi8.com — battle-tested in production.

Affiliate links — they don’t cost you extra and they help keep dibi8.com running.

Last updated: 2026-08-02

References & Sources #

• ZeroClaw
• ZeroClaw homepage
• Ollama