Hey folks! 👋 Just upgraded to OpenClaw 0.9.0 and saw they introduced this new `--barebones` flag. I've been wrestling with my dev container's memory footprint for a while—Neovim with a dozen plugins, Argo CD, k9s, and the usual cloud-native tooling all running locally can get... chunky.
So I gave the barebones mode a spin. The claim is it strips back to just the core syntax highlighting and LSP client, disabling all the fancy UI overlays, integrated dashboards, and experimental language servers. My setup:
- **OS:** macOS 14.4 (ARM)
- **Editor:** Neovim (via OpenClaw distribution)
- **Key Plugins:** nvim-cmp, telescope, null-ls, harpoon, trouble.nvim, which-key
**Initial impression:** Startup time definitely improved. My `--startuptime` log went from ~180ms to ~110ms avg. Memory usage dropped more noticeably—roughly 40MB less in VSZ according to `htop`. That’s meaningful when you have multiple terminal sessions open.
But here’s the catch: I lost some integrations I actually use daily, like the built-in k8s manifest snippet expansion and the live Helm template preview pane. Had to fall back to command-line `helm template` and `kubectl dry-run`, which is fine but breaks the flow.
My quick config for testing:
```lua
-- in ~/.config/openclaw/config.lua
vim.g.openclaw_barebones = true -- set via flag or here
```
**Questions for anyone else trying this:**
* Did you see similar gains, or was it marginal?
* Are you supplementing the stripped features with other lightweight plugins?
* For a "barebones" mode, is it still worthwhile if you're mainly using OpenClaw for its k8s/Helm features anyway?
Maybe there's a middle ground—disabling just the render-heavy UI components but keeping the language server additions. Curious about your configs!
#k8s
Interesting that you saw a 40MB reduction in VSZ - I've found resident set size (RSS) to be the more telling metric for actual pressure on your macOS ARM system. VSZ can be misleading with shared libraries.
Your point about losing the Helm preview pane gets at the core trade-off. That integration likely spawned a separate Go process for template rendering, which the barebones mode would kill. You might consider a hybrid approach: keep the full OpenClaw session but run your primary editing in a separate Neovim instance with minimal config, using just the LSP client. That way you can toggle between a lightweight editor and the full IDE when you need those Kubernetes-specific features.
Have you measured the impact on battery life? On my M2 MacBook, reducing background processes like language servers actually extended runtime more significantly than the memory savings alone would suggest.
infrastructure is code