Moltbot vs OpenClaw: What Changed and How to Migrate

OpenClaw hit 347,000 GitHub stars in under five months, making it the most-starred software repository in GitHub history by March 2026. But the project's three-day triple rebrand, from Clawdbot to Moltbot to OpenClaw between January 27 and January 30, 2026, left thousands of users searching for answers about whether their configs still work.

Moltbot, Clawdbot, and OpenClaw are the same open-source AI agent platform at different stages. Peter Steinberger, the PSPDFKit founder, started it as a weekend side project called "WhatsApp Relay" in November 2025. He named it Clawdbot, a play on Anthropic's Claude model, complete with a lobster mascot. The project grew from a personal AI assistant into a full agent platform with a skills marketplace, multi-channel messaging across 13+ platforms including WhatsApp, Telegram, Discord, Slack, and Signal, persistent memory across sessions, and browser automation.

By April 2026, adoption numbers reflected how fast the project had taken off: 3.2 million monthly active users, 500,000+ running instances across 82 countries, and 44,000+ community-built skills on the ClawHub marketplace. The community around the project grew to 60,000 Discord members and 450,000+ Reddit subscribers.

If you're searching for "moltbot vs openclaw" or "is moltbot the same as openclaw," the answer is yes. Moltbot, Clawdbot, and OpenClaw are the same open-source AI agent platform. The project rebranded from Clawdbot to Moltbot on January 27, 2026 due to trademark concerns, then to OpenClaw on January 30 following a critical security vulnerability. The underlying agent logic, skills API, and architecture are identical across all three names.

This guide covers what triggered each rename, what technically changed, how to migrate an existing Moltbot or Clawdbot installation, and how to set up persistent file storage for your agents after the switch.

The first rename happened on January 27, 2026. Anthropic's legal team contacted Steinberger about the name "Clawdbot." The name sounded too much like "Claude," and the lobster mascot played on Anthropic's branding. Steinberger chose "Moltbot" as the replacement. "Molt fits perfectly," he wrote on the project's blog. "It's what lobsters do to grow." The community updated their installations and moved on.

Three days later, on January 30, Moltbot became OpenClaw. Two simultaneous crises forced the second rename. First, security researchers disclosed CVE-2026-25253, a zero-click WebSocket hijacking vulnerability with a CVSS score of 8.8. Any malicious website could take over a running Moltbot instance by exploiting the WebSocket connection, giving attackers full control of the agent without any user interaction. Second, Snyk discovered what became known as the "ClawHavoc" supply chain attack: over 1,400 malicious skills had been uploaded to MoltHub, the project's skill registry, potentially compromising any instance that installed them.

The OpenClaw rebrand addressed both problems directly. The new release patched the WebSocket vulnerability, introduced cryptographically signed skill manifests to prevent future supply chain attacks, and added eBPF kernel enforcement for skill sandboxing. The old MoltHub registry was replaced with ClawHub, which requires signed manifests for all new skill submissions.

SecurityScorecard later found 135,000 exposed OpenClaw instances across 82 countries, a number that underscored both the platform's massive adoption and the urgency of the security hardening. If you're running an older Moltbot or Clawdbot installation, migrating to OpenClaw isn't just a naming convenience. The security patches alone make it worth the ten minutes.

The core agent logic didn't change during the rebrands. Your skills, prompts, and model configurations carry over. But several things moved, and new capabilities arrived with the OpenClaw release.

Package and CLI changes

The npm package was renamed with each transition, moving from its original Clawdbot name through the Moltbot interim to the current OpenClaw package. The terminal command also changed at each step, so any shell scripts, cron jobs, or systemd services that invoke the old command need updating. Check the official OpenClaw install docs for the current package and command names.

Configuration directory

The default config directory moved with each rename. The current OpenClaw directory stores your main config file, memory files, and custom skill configurations. The built-in migration tool handles the directory move automatically, but manual installations require copying these files. See the migration docs for the exact paths.

Skill registry

MoltHub became ClawHub. The registry URL changed, and skills now require cryptographically signed manifests. Unsigned skills still work when loaded locally, but they can't be installed from the public registry without signature verification.

Model provider configuration

Clawdbot defaulted to Claude without requiring an explicit provider declaration in the config file. OpenClaw requires you to specify your model provider explicitly. If your config file doesn't include a provider section, the agent won't start. This change prevents silent fallback to a provider you didn't intend to use.

New capabilities in OpenClaw

The rebrand coincided with feature additions that go beyond the name change:

• Thinking modes that expose the agent's reasoning process, useful for debugging complex multi-step tasks
• Model fallback chains that automatically switch to a backup LLM provider when the primary is down or rate-limited
• Session isolation for running parallel agents without shared state conflicts
• Sandbox execution that runs untrusted skills in a restricted environment
• Permission scopes and audit logging for tracking what each skill accessed
• Docker deployment support with official Compose files

What stayed the same

The skills API, agent interaction model, model provider support (Anthropic Claude, OpenAI GPT, local models via Ollama), and multi-channel messaging all work identically. If your agent talked to Telegram and Discord under Moltbot, it still does under OpenClaw without changes beyond the package and config path.

Give your OpenClaw agents persistent storage

Free 50GB workspace with MCP access at /mcp. No credit card, no expiration. Your agents read, write, and share files in workspaces that survive restarts and session resets.

Sign Up Free

The migration takes about ten minutes for a typical single-agent setup. Larger deployments with custom skills or Docker infrastructure may need additional time to update references across the stack.

1. Back up your current config and memory

Before changing anything, copy your existing configuration directory and memory files to a safe backup location. If you're migrating from Clawdbot, back up the Clawdbot config directory instead of the Moltbot one. A clean backup lets you roll back if anything breaks.

2. Install OpenClaw

Install the current OpenClaw package globally using npm. The project's README and official docs list the exact package name, which has changed with each rebrand. Keep the old package installed until you've confirmed the migration works.

3. Run the built-in migration tool

OpenClaw includes a built-in migration command that handles the heavy lifting. Point it at your old config file and it copies and reformats the settings, relocates memory files, and rewrites skill paths to point at ClawHub instead of MoltHub. Check the project README for the current command syntax. Read the output carefully. It flags deprecated settings, unsigned skills that need reinstalling, and any config keys that changed names.

4. Update references in your codebase

Search for remaining references to the old names. Common places to check:

• Shell scripts and cron jobs that call the old CLI command
• Custom skill files that import the old tool class names (renamed in OpenClaw)
• Docker Compose files or Dockerfiles with old image names
• CI/CD pipelines that install or invoke the old package
• Environment variables pointing to old config paths

A case-insensitive text search across your project directory catches most of them.

5. Add explicit model provider settings

If your original config relied on Clawdbot's implicit Claude default, you need to declare the provider explicitly. Add a provider block to your config file specifying your preferred LLM provider and model. OpenClaw supports Anthropic Claude, OpenAI GPT, KIMI, local models through Ollama, and others. Check the official configuration docs for the exact YAML format.

6. Verify and test

Run the version check command to confirm the installation. Then test a prompt to verify your agent responds, skills load from ClawHub, and channel integrations reconnect properly.

The most common migration issue is a missing provider declaration, which causes the agent to refuse to start. Check your config file for the provider section first. The second most common problem is unsigned skills that were installed from MoltHub before the ClawHub transition. Reinstall those skills from ClawHub, where they need signed manifests.

Once your OpenClaw instance is running on the new package, the next thing to sort out is where your agent stores its output. OpenClaw agents generate reports, datasets, code artifacts, and media files that need to survive across sessions. Local filesystem storage works during development, but files disappear if you restart the container, and there's no built-in way for a human team member to browse or search what the agent produced.

Typical options for agent file storage each have tradeoffs:

• Local disk: Fast reads and writes, but ephemeral in containerized setups and impossible to share without extra tooling
• S3 or similar object storage: Durable and affordable, but you manage buckets, IAM policies, and access keys. No file previews, no semantic search, no collaboration layer
• Google Drive or Dropbox: Built for humans clicking through a web UI. API rate limits and OAuth complexity make them awkward for agents calling storage operations programmatically

Fast.io fills a different role. It provides workspaces where agents and humans collaborate on the same files, with built-in AI indexing, semantic search, and branded sharing. Your OpenClaw agent connects through the Fast.io MCP server at /mcp, authenticates once via browser, and then creates workspaces, uploads files, and queries documents programmatically.

To connect OpenClaw to Fast.io, install the Fast.io skill from ClawHub. This registers workspace, storage, sharing, and AI tools in your OpenClaw instance. Once installed, your agent can create organized workspaces for different projects, upload files with browser-previewable links for PDFs, videos, images, and code, and enable Intelligence Mode to automatically index every file for semantic search and RAG-powered Q&A with citations.

Branded shares let your agent package deliverables and send them to clients through a professional portal. The ownership transfer feature is especially useful for agencies and consultants: an OpenClaw agent builds a workspace, fills it with deliverables, and then transfers the entire org to the client. The agent retains admin access for ongoing maintenance while the client owns their data.

The free agent plan includes 50GB of storage, 5,000 credits per month, 5 workspaces, and 50 shares. No credit card, no trial period, no auto-deletion. For teams running parallel OpenClaw agents on the same project, file locks prevent write conflicts and webhooks trigger downstream workflows when files change, so you can build reactive pipelines without polling.