Document menu-extra Memory section (active vs. reclaimable cache)

[ronaldmannak]

Summary

Keeps the Install and Run getting-started page in sync with the menu-bar memory change merged in ronaldmannak/PicoServer since the last documentation run. Grounded in the current source (PicoServer/Views/MenuExtraView.swift, Views/Dashboard/MemoryGauge.swift), not a point-in-time PR description.

What changed

• Expanded the menu-extra bullet (step 6) to document the Memory section it shows when the server is running. The usage-bar legend now breaks memory into four buckets — Apps, MLX (active), Cache (reclaimable MLX memory, shown only when there is any), and Free — plus a memory Pressure indicator and a GB / % unit toggle. This reflects PicoServer #51, which split the single MLX bar/legend into active vs. reclaimable cache.
• Before this, master's docs described the menu extra as only status, Start/Stop, and the address + copy button; the Memory section was undocumented on master.

Evidence

$ python3 .claude/skills/write-software-docs/scripts/check_markdown.py --strict \
    docs/user-guide/getting-started/install-and-run-pico-ai-server.md
Checked 1 file(s): 0 error(s), 0 warning(s)

Verified against ronaldmannak/PicoServer master at HEAD (7d67784, includes the #51 merge): MenuExtraView.memorySection's legend rows (Apps blue, MLX pink, Cache pink-at-0.4-opacity gated on mlxCacheMemoryUsed > 0, Free), the GB/% memoryLegendShowsPercent toggle, the Pressure: indicator, and SegmentedMemoryBar's four-bucket (usedNonMLX/mlxActive/mlxCache/free) allocation.

New PicoServer PRs reviewed this run (merged 2026-07-29)

• #51 — split MLX memory bar/legend into active vs reclaimable cache → documented here (the only user-facing surface).
• #61 — freeze Ready page model name to the started download — onboarding Ready-page correctness fix; no documented behavior changes (the Ready page simply shows the correct downloading model). No doc change.
• #67 — don't let notification authorization block an access-narrowing restart — correctness fix so narrowing access (LAN → local-only) restarts the server promptly; the documented bind behavior is unchanged, just made reliable. No doc change.
• #54 — fix HTTPServerActor.startServer reentrancy race — internal concurrency fix, no user/admin/developer surface. No doc change.

PicoCore and PicoDocs had no PRs merged in this window (newest: PicoCore #68 on 07-24, already covered/deferred by the prior run; PicoDocs #40 in June).

Not done on purpose

• Menu-extra Inference section (PicoServer #34) — still not documented. In release builds inferenceProvider is nil, so the Inference section never renders for end users; documenting a DEBUG-only, placeholder-data section would be inaccurate.
• Fuller "Models in Memory" / "Downloads" menu detail — left out to keep this diff scoped to the #51 change; the fuller menu-bar walkthrough is the subject of the separate open PR Add "Tour the Menu Bar" page and correct the onboarding wizard steps #7.

Relationship to the other open documentation PRs

The routine opens one branch/PR per run, so several overlapping menu-bar/onboarding doc PRs are open against master awaiting human review (#4 external MCP servers, #5/#6 onboarding, #7 "Tour the Menu Bar"). All of them predate PicoServer #51, so each still describes the memory legend as Apps / MLX / Free — none reflects the reclaimable-Cache split. This PR is the only one carrying the current four-way legend. If a maintainer merges #7's dedicated menu-bar page, its Memory paragraph should be updated to add the Cache bucket (or this PR's wording grafted in); the two are otherwise equivalent.

---

Generated by Claude Code

[gemini-code-assist]

Caution

The consumer version of Gemini Code Assist on GitHub has been sunset. All code review activity has officially ceased.