Capture what makes you better

Projects go to Linear, team knowledge goes to Notion. What you learn with AI Agents goes to Spark.

Download for macOS GitHub

Saved by your Agent via MCP · 2026-10-14

Why our cache kept serving stale data

Problem

Users saw old prices for minutes after an update, even though the cache TTL was only 30 seconds.

What we learned

A second cache layer in the CDN was ignoring our headers and holding responses for 10 minutes.

Takeaways

Map every cache layer before tuning one
Invalidate on write, don't rely on TTL alone

Cache invalidation patterns · Debugging with an Agent

Spark AI

What did I learn about caching last month?

Two things: check every cache layer, including the CDN, before tuning TTLs¹, and invalidate on write instead of waiting for expiry².

1 Why our cache kept serving stale data2 Cache invalidation patterns

Ask Lulu Spark…

Only what's worth growing on

Technical lessons, conversation conclusions, ideas worth revisiting stay. Status updates and to-dos don't.

Your Agent can save on its own

Two steps, and your Agent saves what's worth keeping on its own.

Or save anything yourself, anytime.

1Install the skill

Teaches your Agent what to save to Spark, and when.

View skill on GitHub →

2Connect MCP

"lulu-spark": {
  "url": "http://127.0.0.1:9876/mcp/<client>",
  "headers": { "Authorization": "Bearer ${SPARK_MCP_TOKEN}" }
}

Any Agent can recall it

Agents search Spark via MCP, so what you learned becomes the next starting point.

Collaborate→Save→Recall↺

Ask Spark AI

Spark AI answers from your notes and helps you tidy them. Every answer links to its source.

Spark AI

Merge my onboarding notes into one page

Done. Merged 3 notes into one page. Removed duplicates and kept links to the originals.

Onboarding.md

Plain Markdown, on your disk

Every note is a .md file in a folder you choose. No lock-in.

Let every session make you better.