docs(wren): document macOS memory first-run scan (#2354)

core/wren/docs/cli.md

@@ -82,6 +82,8 @@ pip install 'wrenai[main]'   # includes memory, interactive, ui
 
 All `memory` subcommands accept `--path DIR` to override the default storage location (`~/.wren/memory/`).
 
+> **Note:** The `memory` extra bundles ~800MB of large unsigned native libraries (lancedb plus sentence-transformers/torch). On macOS, the first command that loads the memory stack can trigger a one-time XProtect/Gatekeeper scan and pause for up to about a minute before it finishes; this is normal macOS behavior, not a Wren error, and happens once per install or fresh virtual environment. With lazy memory loading, lightweight non-`memory` commands are unaffected — the scan is deferred to your first real memory use, not eliminated.
+
 ### Hybrid strategy: full text vs. embedding search
 
 When providing schema context to an LLM, there is a trade-off:

docs/core/get_started/quickstart.md

@@ -235,6 +235,8 @@ Claude Code will:
 7. **Validate and build** — `wren context validate` → `wren context build`
 8. **Index memory** — `wren memory index` (generates seed NL-SQL examples)
 
+> **Tip:** If `wren memory index` (the indexing step above) seems to hang for tens of seconds on macOS — it hasn't. That first `wren memory` command loads large unsigned native libraries (lancedb and torch, ~800MB), and macOS runs a one-time XProtect security scan the first time they execute. This is expected macOS behavior, not a Wren problem, and it's a one-off — every later `wren memory` command runs at normal speed. To avoid the pause during a live demo, run any `wren memory` command once right after install and let it finish.
+
 After completion, verify the project:
 
 ```bash

docs/core/reference/cli.md

@@ -124,6 +124,8 @@ pip install 'wrenai[memory,main]'
 
 All `memory` subcommands accept `--path DIR` to override the default storage location (`~/.wren/memory/`).
 
+> **Note:** The `memory` extra bundles ~800MB of large unsigned native libraries (lancedb plus sentence-transformers/torch). On macOS, the first command that loads the memory stack can trigger a one-time XProtect/Gatekeeper scan and pause for up to about a minute before it finishes; this is normal macOS behavior, not a Wren error, and happens once per install or fresh virtual environment. With lazy memory loading, lightweight non-`memory` commands are unaffected — the scan is deferred to your first real memory use, not eliminated.
+
 ### Hybrid strategy: full text vs. embedding search
 
 When providing schema context to an LLM, there is a trade-off: