OpenClaw is a personal AI assistant runtime that treats chat channels as the interface, not as an afterthought. The center of the project is a local Gateway that owns sessions, model routing, tools, channel adapters, companion apps, and a Canvas surface. The README lists a long channel matrix, but the more useful way to think about the repo is narrower: it is for people who want one agent identity reachable from their existing messaging apps while still keeping the control plane on their own machines.

That framing matters because OpenClaw is not a generic automation builder. It overlaps with workflow tools, agent frameworks, desktop assistants, and self-hosted chat UIs, but its bet is different. It wants the assistant to live where a user already talks: Telegram, WhatsApp, Slack, Discord, Google Chat, Signal, iMessage, Microsoft Teams, Matrix, LINE, WeChat, QQ, WebChat, and several other channels. It also ships optional macOS, iOS, Android, and Windows surfaces. If you are evaluating it, the question is less “can it run an agent?” and more “do I want an always-on Gateway that can receive real messages and act through local tools?”

The growth number deserves care. The repository was created in November 2025 and had 378,109 stars as of 2026-06. Its sampled star history shows a jump from 7,001 stars on 2026-01-24 to 39,901 stars on 2026-01-26, then 378,109 by 2026-06-11. That is an unusual curve for infrastructure software. It does not make the project fake, and it does not prove adoption by itself. It does mean you should read the docs, issues, release cadence, and security model before treating the star count as a maturity signal.

What OpenClaw Actually Gives You

The Gateway is the durable part. It runs locally, keeps sessions, routes model calls, brokers channel traffic, and exposes tools. The README calls the Gateway the control plane, and that is accurate: the product is the assistant experience, but most operational risk sits in the Gateway.

The channel surface is broad. OpenClaw documents support for mainstream chat apps, team chat, IRC, Matrix, Nextcloud Talk, Nostr, Twitch, Zalo, Tlon, WeChat, QQ, and WebChat. That list is valuable only if you need cross-channel continuity. If your main need is one web chat with a knowledge base, the channel count becomes overhead.

The local apps are optional. macOS and Windows companion apps add tray or menu bar control, local node behavior, chat, voice, and Canvas-related features. iOS and Android nodes pair with the Gateway over WebSocket. This is useful for a personal assistant, but it also creates a wider permission surface than a plain hosted chatbot.

OpenClaw also leans into tools and skills. The README points to browser, canvas, nodes, cron, sessions, Discord and Slack actions, plus a skills system and ClawHub registry. That makes the repo interesting for operators who want an agent to do real work, not only produce text.

Install And First Run

The README states that OpenClaw needs Node 24, or Node 22.19 and newer. It says the preferred setup is openclaw onboard, with npm, pnpm, or bun support. The recommended install path is:

npm install -g openclaw@latest
openclaw onboard

The README also lists pnpm as an alternative:

pnpm add -g openclaw@latest
openclaw onboard

After daemon setup, the quick check is:

openclaw gateway status

The README also documents daemon installation, foreground Gateway debugging, direct message sending, and agent prompts. Those examples show the product shape: the Gateway receives or sends channel messages, then an agent session can respond with a model and tools behind it. Use the README for the exact flags, since those options can change faster than the editorial judgment on this page.

Security Model: Read This Before Remote Exposure

OpenClaw connects an agent to real messaging surfaces. The README explicitly says inbound DMs should be treated as untrusted input. That is the correct default assumption. A user who connects Telegram, WhatsApp, Signal, iMessage, Teams, Discord, Slack, or Google Chat is exposing the assistant to people and group contexts that may not share the operator’s intent.

The default DM policy in the README is pairing-based for major DM-capable channels. Unknown senders receive a short pairing code, and the assistant does not process their message until the operator approves it:

openclaw pairing approve <channel> <code>

The public inbound mode is an explicit opt-in. The README says it requires dmPolicy="open" and an allowlist wildcard in channel config. That is the line between a personal assistant and a remotely reachable automation endpoint. Do not cross it casually.

OpenClaw’s own VISION file says the current priorities include security and safe defaults, bug fixes and stability, and setup reliability. That matches the open issue surface as of 2026-06: many current reports are about message loss, provider authentication, session state, reply routing, and security boundaries. These are the right areas to watch for a project that can talk to real people and invoke local capabilities.

Where It Fits

Use OpenClaw if you want a single personal assistant that can live in your chat channels, run on your devices, and talk to local tools. It is especially interesting if you care about Telegram, WhatsApp, Slack, Discord, iMessage, or Matrix as first-class surfaces rather than notification targets.

Use it cautiously if your use case needs remote access. The docs include a security section and an exposure runbook because remote Gateway access is not a cosmetic deployment choice. It changes the threat model.

Skip it if your goal is deterministic workflow automation with a visual builder. n8n is a better fit for scheduled integrations, enterprise workflows, and explicit nodes. Dify is closer when the problem is building and operating agentic applications or model-backed workflows behind a product surface. AutoGen is closer when you want a Python framework for agent coordination in your own application code.

Also skip it if you only want a polished chat UI. LobeHub and other chat-oriented projects usually present a narrower, easier operator surface. OpenClaw’s value is the Gateway plus channels plus tools. Without that need, the setup weight is hard to justify.

Alternatives Compared

Project Stars as of 2026-06 Language License field Better fit
OpenClaw 378,109 TypeScript MIT in LICENSE, API reports NOASSERTION Personal assistant Gateway across real chat channels
n8n 192,025 TypeScript NOASSERTION Visual workflow automation and integrations
Dify 144,836 TypeScript NOASSERTION Product-facing agentic workflow development
LobeHub 78,504 TypeScript NOASSERTION Agent operations and chat-oriented assistant surface
AutoGen 58,870 Python CC-BY-4.0 Programmable multi-agent framework in Python

This comparison is intentionally not a winner table. The projects sit near one another in AI automation search results, but their operating models differ. OpenClaw is the most channel-native of the group. n8n is the clearest workflow product. Dify is closer to app teams building agent workflows. AutoGen is a framework, not a personal assistant runtime.

What The Issues Say

The open issue list is unusually active. That is partly expected for a young, popular project with many integrations, but the topics matter. Recent issues as of 2026-06 include DeepSeek provider routing going through the wrong OpenAI-style transport, Telegram explicit sends failing with an unsupported-channel error, Telegram spooled updates being deleted after turn failure, session yield completion being easy to ignore, iMessage reply metadata crossing a private-API boundary, and a WhatsApp identity confusion report with security labels.

You do not need to panic over every issue title. You do need to understand the pattern. OpenClaw lives at the boundary between LLM sessions, provider credentials, messaging identity, and host tools. Bugs there can turn into message loss, wrong-recipient delivery, credential boundary problems, or stuck sessions. For hobby use, that may be acceptable. For anything involving customers, private work channels, or production automations, test the exact channel and provider path before you depend on it.

The PR queue also shows fast-moving maintenance. In one recent snapshot, open PRs covered Telegram spool handling, QQBot private command gates, custom node IDs, Responses system prompts, cron status, and UI layout. Fast maintenance is good, but it also means the stable behavior may shift quickly from release to release.

Growth Curve Reading

The star curve is steep and sparse. The sampled history has only ten points, so it should not be overread. Still, the shape is clear enough to say this: OpenClaw did not grow like a slow infrastructure library. It moved from a small project to a very large GitHub signal in weeks, then kept climbing.

That makes the page’s live data block more useful than a static endorsement. Watch forks, issue count, recent commits, and releases together. A repo with 378,109 stars and 8,030 open issues as of 2026-06 may have strong community energy, maintenance pressure, or both. The safe reading is that OpenClaw has attention and momentum, while its operational maturity still needs case-by-case verification.

If the OpenClaw channel-first model feels too broad, compare it with n8n for workflow automation. For a broader developer tool baseline, microsoft/vscode gives a useful contrast: a mature extensible product with a long release and extension ecosystem. For curated learning rather than an installable assistant, public-apis is a different kind of GitHub asset, but it shows how a large repository can be valuable without running any local daemon.

FAQ

Is OpenClaw ready for production use?

Not as a blanket statement. It is active, has releases, and documents security defaults, but it is very young and has many current issues around message delivery, provider routing, session state, and security boundaries as of 2026-06. Treat it as a project to pilot carefully, not as a drop-in production control plane.

How do I install OpenClaw?

The README recommends Node 24 or Node 22.19 and newer, then npm install -g openclaw@latest followed by openclaw onboard. pnpm global install is also documented. After that, run openclaw gateway status. The README documents the daemon installation flag for users who want the Gateway to stay running.

Is OpenClaw the same kind of tool as n8n?

No. OpenClaw is a personal AI assistant Gateway for chat channels and local tools. n8n is a visual workflow automation platform. They can overlap around automation, but the operator model is different.

Does OpenClaw support Telegram and WhatsApp?

The README lists both Telegram and WhatsApp among supported channels. For real use, verify the exact channel path you need, because current open issues include Telegram send and spool problems, and one WhatsApp identity confusion report carried security labels as of 2026-06.

What should I check before exposing OpenClaw remotely?

Read the Gateway security docs and exposure runbook, keep DM pairing enabled unless you have a clear reason, run openclaw doctor, and test allowlists with a non-sensitive account first. Remote exposure changes OpenClaw from a local assistant into a reachable automation surface.

Why is the star count so high for such a young repo?

The repo was created in November 2025 and had 378,109 stars as of 2026-06. The sampled star history shows very fast jumps in January 2026 and continued growth afterward. That is a signal of attention, but not proof of maturity or safe deployment.