Technology · Agentic → Agentic Gateways · Open Source

OpenClaw

Taxonomy

OpenClaw is a self-hosted gateway that puts messaging channels (WhatsApp, Telegram, Signal, iMessage, Slack, Discord, Teams, Matrix, …) in front of an agent on your machine, with allowlists, pairing, heartbeats, and exec policy. It is not a frontier model and not a second Claude Code. The brain is whichever backend you configure — often Claude API / Claude CLI reuse, or Ollama locally. In this catalog it sits with tech-agentic-gateways: channels + policy shell, not the company ledger.

DocsStack: agentic-agentic-gateways

Why it matters here

Personal agents fail when “chat” and “authority” collapse into one product. OpenClaw’s scarce pieces are standing channels, always-on gateway policy, and an optional local host — not a better Claude. Compare MCP-oriented peers (Agent Gateway, FastMCP, Supergateway) when the bottleneck is tool transport, not phone inbox.

How it works

Messages arrive on a channel. OpenClaw applies allowFrom / pairing / exec rules, then calls the configured model backend (API key, Claude CLI, or Ollama). Workspace files hold persona; restricted jobs can expose only allowlisted tools via grant-scoped MCP instead of the full native tool set.

Related: Claude API · Claude Code · Ollama · Agent Gateway · FastMCP · tech-agentic-gateways · tech-coding-agents

Flow

When to reach for it

  • Use when: you already have a strong coding/model stack and need standing chat channels plus host-side exec policy.
  • Skip when: IDE/CLI coding alone is enough — stay on Claude Code / Continue; or you need MCP registry plumbing without messaging — prefer MCP Registry / Supergateway.
  • Prefer instead: Claude Code for repo work; Ollama when the scarce buy is local weights, not Telegram.

Limits

  • Shell ≠ brain: OpenClaw does not multiply model quality; a weak local model on a Studio is silence, not Opus.
  • Attack surface: messaging front doors + exec policy misconfig = remote command path — start conservative (allowFrom, dedicated number).
  • Scope: gateway policy is not the company approval ledger.

What we checked

Claims below are backed by science sources on disk.

Product / channels

OpenClaw site · STRONG

“The AI that really does things.”

Docs / gateway

OpenClaw docs · STRONG

“Any OS gateway for AI agents across Discord, Google Chat, iMessage, Matrix, Microsoft Teams, Signal, Slack, Telegram, WhatsApp, Zalo, and more.”

GitHub

OpenClaw GitHub · MODERATE

“Self-hosted gateway connecting chat apps to AI agents”
Research inventory

9 tags · 6 out · 7 in · 3 artifacts · 1 gaps · 2 corpus docs

Catalog tags

landscape.layer
Agentic
landscape.subcategory
Agentic Gateways
license_tag
Open Source
maps.dm
absent
maps.features
0
one_liner
Agentic
review.depth
science
slug
openclaw
title
OpenClaw

Artifacts

  • dm_map · absent
  • features_map · present · technologies/openclaw/features.md
  • readme · present · technologies/openclaw/README.md

Out · alternative_to

In · alternative_to

In · in_stack

Corpus tags

kind
company_review
slug
openclaw
surface
interface-review

Corpus documents (2)

company_review · 2

  • OpenClaw · interface review
  • OpenClaw