Adapted from the official Clawdi/Phala Network announcement. Originally published at clawdi.ai.
If you’ve been building with AI agents for a while, you’ve probably run into the same problem: every time you switch frameworks, you lose everything. Your memory, your connections, your scheduled jobs, your custom skills. It all has to be rebuilt from scratch because it was tied to the agent, not to a workspace you actually own.
Clawdi solves this in a way that’s worth understanding properly, because it changes how you think about agent infrastructure entirely.
What’s Actually Happening Under the Hood
When you run OpenClaw or Hermes on Clawdi, the agent itself is not where your data lives. Phala Network’s Intel TDX confidential virtual machine is. That’s the encrypted environment where your MCP servers, memory, skills, connections, and cron jobs are all stored. The agent is just the layer that interacts with them. The workspace is the thing that actually holds your setup together.
Because that workspace runs inside a TEE-secured cloud environment, everything in it stays private by default. No outside access, no exposure at the infrastructure level. It’s not something you have to configure or turn on. It’s just how it works from the start.
Why Switching Frameworks No Longer Means Starting Over
This separation between the agent and the workspace is what makes framework switching painless. Since your memory, tools, and configuration all live in the encrypted workspace rather than inside a specific agent, you can move from OpenClaw to Hermes, or the other way around, without rebuilding anything. Your setup follows the workspace, not the agent you happened to be using.
For developers who are still deciding which framework fits their workflow, or who want the flexibility to change direction later, this is a meaningful shift. You’re not locked in by the infrastructure. You can explore, iterate, and switch without the cost of starting over every time.
If you want to explore Clawdi or try it yourself, head to https://www.clawdi.ai/
Top comments (0)