Open-source AI personal assistant framework by Peter Steinberger. Locally executable, multi-channel, community-driven. 250,000+ GitHub stars in 60 days.
OpenClaw runs locally on your own hardware — or self-hosted in your own data centre. Combined with open-source LLMs (Mistral, Llama 3, DeepSeek) via Ollama, not a single byte of your data leaves your own infrastructure. For government, healthcare, financial services, defence and any regulated environment, OpenClaw is thus the superior alternative to SaaS AI agents.
OpenClaw is an open-source framework for AI personal assistants — developed by Austrian developer Peter Steinberger, previously known in the iOS developer world through PSPDFKit and the "vibe coding" movement. OpenClaw agents are not IDE tools like Claude Code or Cursor — they are standalone assistants reachable across any messaging channel and orchestrating autonomous workflows.
The project history is unusual: first release in November 2025 under the name Clawdbot. Renamed to Moltbot on January 27, 2026, due to trademark complications with Anthropic. Three days later renamed to OpenClaw because Steinberger found the new mark more memorable. Within ~60 days the repo crossed 250,000 GitHub stars, making it one of the fastest- growing open-source projects ever.
In February 2026, Steinberger announced he is joining OpenAI. Stewardship of the project moves to a newly formed non-profit foundation — the roadmap is thus community-secured, independent of a single maintainer.
Current stable release: v2026.4.12 (as of prompt date 2026-04-15). Release cadence is roughly weekly — OpenClaw is production- ready but moves fast as a young OSS project. Production deployments should be coupled with version pinning and a security review process.
One agent, all messengers: WhatsApp, Telegram, Slack, Discord, iMessage, Microsoft Teams, Matrix. Users interact where they already are — no new app needed.
Bots run locally on laptop, VPS, Kubernetes cluster or on-prem. Data leaves your infrastructure only if an external LLM is wired in — and even that is optional via Ollama.
Personality and ethics manifesto of the agent as a declarative markdown file: who it is, how it speaks, boundaries it does not cross. Versionable like code.
Community skill marketplace with ~162 skills across 19 categories. Integration, CRM, scheduling, research, notifications, automation. Analogous to npm or pip, but for agent skills.
Backend LLM is swappable: Claude (Anthropic), GPT (OpenAI), DeepSeek, Mistral, Llama 3 via Ollama. No vendor lock, full model sovereignty.
When run locally + open-source LLM: no DPA needed because no data transfer happens. The ideal architecture for government, healthcare, finance.
Where OpenClaw is the rational choice over SaaS AI agents.
IT consolidation and BSI baseline protection rarely allow SaaS AI. OpenClaw + local LLM is self-hostable, auditable, without third-country transfer. Ideal for citizen services, internal workflows, FAQ bots.
Patient data must not leave the institution. OpenClaw in on-prem deployment with OSS LLM meets Art. 9 GDPR (special categories) and German § 203 StGB.
BaFin / ECB supervision requires data residency + auditability. OpenClaw can be set up as reproducible deployment via Kubernetes + ArgoCD, logs fully transparent.
Craftspeople, service firms, law offices with PII touchpoints (mandate data) benefit from local-first setup. OpenClaw + local Llama 3 = no copy of your client lists at OpenAI or Anthropic.
Research data must remain under institutional control. OpenClaw + Ollama on the university cluster is faster to set up than the typical SaaS DPA and legally clean.
| Criterion | OpenClaw | SaaS agents (Zapier AI, Lindy, etc.) |
|---|---|---|
| Data residency | Own infra / EU | US / hybrid |
| DSGVO / GDPR | By design | AVV nötig |
| Vendor lock | None (OSS) | High |
| Running cost | Infra cost + LLM tokens | Seat pricing + API markup |
| Time to deploy | 1-2 weeks (self-host) | Hours (SaaS) |
| Skill ecosystem | ClawHub (~162) | Proprietary, vendor-curated |
| Auditability | Fully transparent (OSS + own logs) | Vendor logs, no code insight |
Bottom line: SaaS agents are faster to set up. OpenClaw is legally and data sovereign. In regulated markets the latter decides — not setup time.
A community security audit of ClawHub skills found that around 12% of submitted skills can contain malicious code (as of early 2026). Before installing new skills, a code review is mandatory; in production setups ideally automated (SAST, dependency check, sandbox run). Teams rolling out OpenClaw in DACH companies should maintain a skill allow-list.
Anthropic's coding agent in comparison — managed SaaS vs. self-host.
Category overview with DACH providers for OpenClaw deployments.
Overview of all relevant AI dev tools 2026 with positioning.
Official repo with install guide and example SOUL.md.
OpenClaw is an open-source AI personal assistant framework released by Peter Steinberger (Austrian developer) in November 2025 — initially as Clawdbot, briefly as Moltbot (renaming due to trademark complications), and since early 2026 as OpenClaw. The project surpassed 250,000 GitHub stars within ~60 days, making it one of the fastest-growing open-source projects of all time.
Claude Code and Cursor are coding tools in the narrow sense. OpenClaw is a personal assistant framework running across many channels — WhatsApp, Telegram, Slack, Discord, iMessage, Microsoft Teams, Matrix — and uses swappable backend LLMs (Claude, DeepSeek, GPT, local via Ollama). Agents are configured via SOUL.md (personality + ethics) and skills. OpenClaw is thus a foundation for autonomous workflows, not an IDE replacement.
Yes — OpenClaw bots run locally or self-hosted on your own infrastructure. Combined with open-source LLMs (Mistral, Llama 3, DeepSeek) via Ollama or vLLM, a complete system can operate without data leaving your own data centre. This is the central GDPR advantage over SaaS agents and makes OpenClaw interesting for government and regulated sectors.
ClawHub is the OpenClaw community skill marketplace — comparable to NPM or pip, but for agent skills. As of early 2026, ClawHub lists ~162 community skills in 19 categories (per project brief). Caveat: a community security audit found that ~12% of submitted skills contained malicious code — every install should be auditable.
SOUL.md is the central personality and ethics configuration file of an OpenClaw agent. It defines who the agent is, how it behaves, which boundaries it does not cross. The concept combines prompt engineering with declarative policy rules — similar to claude.md in the Claude Code ecosystem, but broader for multi-channel workflows.
Peter Steinberger announced in February 2026 that he is joining OpenAI. Stewardship of OpenClaw is taken over by a non-profit foundation — analogous to models like the Python Software Foundation or the Linux Foundation. The roadmap thus continues community-driven.
OpenClaw is production-ready but young: stable-release cadence is weekly. DACH companies wanting to deploy OpenClaw should engage an agency or freelancer with OSS experience — especially for skill selection (ClawHub security), LLM backend integration and security review. Trusted AI Partners matches verified DACH specialists.
Describe your OpenClaw project — we match DACH agencies and freelancers with OSS AI expertise within 24 hours.