Claw Park is the ecosystem map for the growing Claw family. The goal is simple: when a new Claw project appears, you should be able to answer two questions quickly: what is it trying to do, and where does it fit in the stack?
The Claw ecosystem is no longer one assistant repo. It now spans:
- Gateway and core runtime
- Registries, bundles, and plugin distribution
- Packaging and deployment
- Research workspaces and daily copilots
- Scientific specialists and evolution engines
- Fully autonomous research pipelines
- Desktop reading surfaces and community catalogs
Without a map, all of these look like sibling products. They are not. Claw Park keeps them separated by job, not by branding.
| Project | What it is doing | Role in the stack | Best for |
|---|---|---|---|
| OpenClaw | General assistant platform that is increasingly a gateway, control surface, and plugin runtime | Gateway / foundation | Builders who want the main runtime and ecosystem entry point |
| ClawHub | Public registry for OpenClaw skills and plugins with search, install, publish, and versioning flows | Registry / discovery | Users who want to find, install, and publish capabilities |
| OpenClaw Plugin Bundles | Compatibility layer that maps Codex, Claude, and Cursor bundles into native OpenClaw features | Compatibility layer | Teams that want to reuse third-party agent bundles without rewriting them |
| nix-openclaw | Declarative Nix packaging and deployment path for OpenClaw across macOS and Linux | Packaging / deployment | Operators who care about reproducible installs, rollback, and fleet-style setup |
| InnoClaw | Self-hostable research workspace for grounded chat, paper study, scientific skills, and research execution | Research workspace | Labs and self-hosters who want files, papers, and execution in one place |
| ResearchClaw | Personal research assistant for literature review, note-taking, experiment tracking, and paper writing | Daily research copilot | Individual researchers who want an end-to-end daily driver |
| ResearchClaw Desktop App | Local-first desktop app for PDF chat, note-taking, and paper reading workflows | Desktop / reading surface | Readers who want a lighter-weight local research surface |
| ScienceClaw | Self-evolving scientific research colleague with strong research memory and scientific focus | Scientific specialist | Research-heavy users who want a more opinionated scientific agent |
| ScienceClaw (alt repo) | Lab-style research automation stack with broad database access and multi-agent coverage | Research-lab variant | Users exploring aggressive lab automation patterns |
| MetaClaw | Learning framework that extracts reusable skills from real use and supports online evolution | Evolution engine | Builders interested in adaptive agents and learning-in-the-loop systems |
| AutoResearchClaw | Fully autonomous idea-to-paper pipeline with experiments, review, verification, and final deliverables | Autonomous pipeline | People testing how far autonomous research generation can go |
| awesome-openclaw-skills | Community-maintained directory of OpenClaw-style skills and examples | Community catalog | Users who want fast discovery of skills and ecosystem patterns |
Think in layers instead of sibling products:
| Layer | Main question | Representative pieces |
|---|---|---|
| Gateway / foundation | What runtime and control surface am I building on? | OpenClaw |
| Registry / discovery | Where do skills and plugins get published, versioned, and installed? | ClawHub · awesome-openclaw-skills |
| Compatibility | How do I reuse adjacent agent ecosystems without rewriting everything? | OpenClaw Plugin Bundles |
| Packaging / deployment | How do I install, pin, update, and roll back reliably? | nix-openclaw |
| Research workspace | Where do I chat over files, papers, and tasks? | InnoClaw · ResearchClaw Desktop App |
| Daily research copilot | What helps me read, track, and write every day? | ResearchClaw |
| Specialist scientist | What pushes deeper into scientific assistance and memory? | ScienceClaw |
| Evolution layer | What helps the agent learn from use over time? | MetaClaw |
| Autonomous pipeline | What tries to do the full research loop for me? | AutoResearchClaw |
This avoids a common mistake: comparing ClawHub to ResearchClaw, or nix-openclaw to ScienceClaw, as if they were trying to solve the same problem.
The main reason OpenClaw matters in 2026 Spring is not just popularity. It now reads like a stack rather than a shell:
- Gateway: the core runtime sits between users, chat surfaces, models, tools, and plugins.
- Registry: ClawHub turns skills and plugins into a searchable, versioned public layer.
- Compatibility: Plugin Bundles let OpenClaw reuse Codex, Claude, and Cursor ecosystem formats.
- Deployment path: nix-openclaw makes the ecosystem easier to package, pin, and operate.
That makes OpenClaw more important as ecosystem infrastructure than as just one more assistant app.
The interesting shift is not just "more Claws." It is that research-oriented Claws are splitting into clearly different bets:
| Pattern | Representative projects | What the bet is |
|---|---|---|
| Gateway + ecosystem layer | OpenClaw · ClawHub · Plugin Bundles | The moat is distribution, compatibility, and a reusable runtime surface |
| Reproducible deployment | nix-openclaw | Operators want a standard way to install and maintain the ecosystem across machines |
| Grounded research workspace | InnoClaw · ResearchClaw Desktop App | Researchers want file-aware, paper-aware, local-first workspaces instead of generic chat |
| Daily research copilot | ResearchClaw | The core value is steady literature, notes, tracking, and writing support rather than maximum autonomy |
| Scientific specialist | ScienceClaw · ScienceClaw (alt repo) | The agent should behave more like a persistent scientific collaborator or even a mini research lab |
| Learning / evolution engine | MetaClaw | The long-term moat is online learning, skill extraction, and adaptation from real use |
| Full autonomy pipeline | AutoResearchClaw | The system should run as much of the idea-to-paper loop as possible with minimal intervention |
This is why "Which Claw is best?" is often the wrong question. The better question is which layer of the research stack you are trying to strengthen.
MetaClaw is the clearest Claw-native representative of the learning layer, but the surrounding ecosystem is now broader than Claw itself.
| Outside the Claw family | What it contributes |
|---|---|
| Agent Lightning | General agent training with RL, automatic prompt optimization, and SFT |
| Agent0 · AgentEvolver | Self-generated evolution loops, zero-data improvement, and agent learning from their own exploration |
| EvoAgentX · EvoScientist | Workflow-level evolution and scientist-loop optimization |
| Acontext | Persistent context, memory, and reusable skills as part of agent improvement |
The useful mental model is: MetaClaw shows how learning can live inside the Claw ecosystem, while the wider self-evolving-agent space shows that this is becoming a separate stack layer across the whole agent field.
The broader ecosystem is no longer only project repos. It now includes a distribution and control layer around those repos:
| Layer | Example | Why it matters |
|---|---|---|
| Skill and plugin registry | ClawHub | Makes OpenClaw look like a living ecosystem with versioned, searchable skills and plugins |
| Compatible bundles | OpenClaw Plugin Bundles | Suggests agent ecosystems may interoperate through installable bundle compatibility, not only isolated plugins |
| Deployment substrate | nix-openclaw | Gives operators a standard path for reproducible setup, upgrades, and rollback |
| Skill discovery | awesome-openclaw-skills | Makes OpenClaw-style skills easier to browse, compare, and reuse |
| Chat control surface | cc-connect | Lets teams operate terminal agents from messaging tools instead of requiring everyone to sit inside a shell |
| Tool registry | Official MCP Registry · awesome-mcp-servers | Standardizes discovery and installation of external tools that research agents depend on |
This is not "another Claw," but it changes how Claws spread: through registries, compatibility layers, packaging paths, and remote-control surfaces rather than only monolithic apps.
| If you want... | Start here |
|---|---|
| The main runtime and gateway | OpenClaw |
| A public registry for skills and plugins | ClawHub |
| To reuse Claude / Cursor / Codex bundles | OpenClaw Plugin Bundles |
| Declarative deployment and rollback | nix-openclaw |
| A grounded research workspace | InnoClaw |
| A practical personal research assistant | ResearchClaw |
| A local paper-reading surface | ResearchClaw Desktop App |
| A more research-specialized evolving scientist | ScienceClaw |
| Learning and evolution infrastructure | MetaClaw |
| Full idea-to-paper autonomy | AutoResearchClaw |
- Read Tools & Platforms for the broader research tool stack.
- Read Systems if you care about end-to-end autonomous research.
- Read Vibe Coding if your next question is how these systems actually execute code and repo work.
Home: README · Prev: Tools & Platforms · Next: Vibe Coding