Skip to content

docs: 2026-08-24 performance and efficiency deep dive - #40

Closed
sambitcreate wants to merge 1 commit into
feature/bots-and-iosfrom
cursor/performance-efficiency-audit-7b00
Closed

docs: 2026-08-24 performance and efficiency deep dive#40
sambitcreate wants to merge 1 commit into
feature/bots-and-iosfrom
cursor/performance-efficiency-audit-7b00

Conversation

@sambitcreate

Copy link
Copy Markdown
Owner

Five parallel source audits of Aiden’s performance and efficiency, then a line-checked current-state document.

What this is

A read-only deep dive saved as docs/performance-efficiency-audit.md. It does not implement the master plan. It records what is true on 94e612a so the July 2026 plan is not used as if nothing has changed.

The existing plan is still the delivery vehicle. This PR only:

Verdict

The expensive parts are still compounding always-on or unbounded paths, not Electron vs React. Highest remaining leverage:

  1. Stop embedding historical attachment bytes in chat JSON/IPC, and stop pretty-rewriting the whole transcript on every append
  2. Stop idle Git subprocesses (~6 commands every 5s in a folder chat) and the perpetual streaming-reveal RAF loop
  3. Move Parakeet off the main thread and cap voice IPC
  4. Make parent coding-tool Stop real; own MCP close, crash recovery, and model install
  5. Then renderer scale, startup split, map-free packages, and a real measurement loop

What already landed (do not re-open as P0)

  • Atomic chat/index writes
  • Attachment ingest count/byte caps and sequential reads
  • Main-owned generation history (no renderer-supplied base64 history for generate)
  • MCP connect single-flight

What the July plan missed

Telegram multi-profile long-poll, Aiden Remote SSE + iOS full-chat reloads, and subagent utilityProcess cost are now first-class efficiency surfaces. iOS itself is largely cache-first and bounded.

Packaged JS size, source-map megabytes, and energy numbers were not remeasured in this environment. Config still ships maps and an eager renderer graph.

Testing

Docs only. No runtime behavior change.

Open in Web Open in Cursor 

Capture a current-state audit of idle work, memory bounds, renderer
scale, packaging, and iOS/Remote/Telegram cost. Point the existing
performance plan at the new evidence so July claims that already
landed are not treated as open.

Co-authored-by: Sambit Biswas <sambitcreate@users.noreply.github.com>
@sambitcreate

Copy link
Copy Markdown
Owner Author

Closing because this draft targets the superseded feature/bots-and-ios base after the consolidated mobile/Bots work moved to #41. The source branch is intentionally preserved because its research documents are not included in #41.

@sambitcreate
sambitcreate deleted the cursor/performance-efficiency-audit-7b00 branch August 25, 2026 14:35
@sambitcreate
sambitcreate restored the cursor/performance-efficiency-audit-7b00 branch August 25, 2026 14:38
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants