Small, focused change. The pressure chip on PipelineMemoryGraph and the badge bump on the Performance tab when memory pressure is high or critical are both the right shape — wiring global.memory_pressure through as an optional prop keeps the graph component reusable, and the pressureChips lookup table is a clean way to map the four pressure levels to display label + Tailwind class. low resolving to a hidden chip rather than a conditional {#if} is a fine tradeoff.

A few small things:

• runtimeErrorsCount = connectorsWithErrorsCount + memoryPressureErrorCount conflates two different signals into a single integer badge. A user looking at a "2" badge can't tell whether that's "2 connector errors" or "1 connector + memory pressure." For now this is fine because the breakdown is visible inside the panel, but if a third "runtime issue" source ever appears the badge becomes a meaningless count. Worth a comment noting the badge is a "has issues" indicator and the number is approximate.
• The pressureChips record uses 'bg-error-50-950' for both high and critical. Visually identical for the user; the only distinction comes from the label text. Was that intentional? If critical is meaningfully more severe (e.g., imminent OOM) than high, a stronger color or a pulsing border would help users triage at a glance.
• The .devcontainer/devcontainer.json flip from "editor.formatOnSave": true to false is unrelated to memory pressure and probably belongs in a separate PR — easy to miss in review. Not blocking, just noting it.
• Type import comes from $lib/services/manager while the existing pipeline type comes from $lib/services/pipelineManager — slightly confusing aliasing, but consistent with what's already there.
• The multihostMemoryLimitMb change is pure formatting (adding parens around the conditional expression). Biome reflow — fine.

mihaibudiu already approved. I won't block on the nits above. APPROVE.