Chunking Settings Explained

Most AI models can only "see" a few thousand words at a time — that's their context window. When Time Garden tries to summarize a whole month or year, the raw text far exceeds that limit.

Chunking is the trick that solves this: instead of sending one giant blob to the model, Time Garden splits it into pieces, summarizes each piece, then synthesizes the summaries.

---

When Chunking Matters

So for weekly and up, chunking kicks in. For dailies it doesn't.

---

How Chunking Works In Time Garden

When you click "Weekly Summary" and chunking is enabled:

1. Time Garden takes each of the 7 daily entries
2. Sends each one to the chunking model with a prompt like "summarize this day in N words"
3. Collects those 7 mini-summaries
4. Sends them all together to the summary model for final synthesis

Same idea for monthly, but it can chunk either the daily notes (slow, granular) or the weekly summaries (fast, broader). You choose.

---

The Settings

In Obsidian Settings → Time Garden Plugin → Chunking you'll find:

enabled

Toggle chunking globally. Default: on. Turn it off if you've moved to a model with a massive context window (e.g., 100K+ tokens) and want to skip the chunk-then-synthesize step.

detailedWordCount (default: 150)

When chunking generates a "detailed" summary of a daily entry, aim for this many words. Higher = more detail preserved, larger total input to the synthesis step.

conciseWordCount (default: 50)

The "concise" version of each daily summary, used when context budget is tight.

weeklyDetailedWordCount / weeklyConciseWordCount

Same idea, but for chunking weekly summaries (used when generating monthly+ summaries).

useDailyChunksForMonth (default: on)

When generating the monthly summary, include daily chunks. More granular monthly summaries.

useWeeklyChunksForMonth (default: on)

Also include weekly summaries in the monthly summary input. Combines well with the option above.

useWeeklyChunksForQuarter (default: on)

Use weekly summaries when generating the quarterly summary.

useDailyChunksForYear (default: off)

Use daily chunks when generating the yearly summary. Off by default — 365 chunks is a lot, and the yearly synthesis usually works fine from just quarterly summaries.

maxTokensPerOperation (default: 4000)

A hard cap on total tokens fed into any single AI call. If your chunks would exceed this, Time Garden truncates the lowest-priority ones first. Increase if you have a model with a bigger context window.

chunkingModel

Which model does the chunking step itself. Defaults to phi4-mini because chunking is many short operations — speed matters more than depth here.

---

Common Tuning Scenarios

"My monthly summary is too vague"

Try:

• useDailyChunksForMonth = on (if not already)
• Increase detailedWordCount to 200
• Upgrade your summary model for the month operation in Choosing Your AI Models

"My quarterly summary takes 8 minutes"

Try:

• useDailyChunksForMonth and useWeeklyChunksForMonth = on, but useDailyChunksForYear = off (already default)
• Drop your monthly summary chunking model to phi4-mini if it's heavier
• Check maxTokensPerOperation — lower it if your model is small

"Summaries leave out important details"

Try:

• Increase detailedWordCount and conciseWordCount
• Turn on Include Highlights / Progress in Customizing AI Prompts (these don't go through chunking — they're appended directly)

"Everything is fast but generic"

The cause is rarely chunking — it's usually the summary model. Try a smarter one for the period that feels generic. (See Choosing Your AI Models.)

---

A Mental Model

Think of chunking as the librarian before the AI author. The librarian skims every book and writes a 1-page synopsis. The author then writes a 5-page review only from the synopses.

You can:

• Make the librarian write longer synopses (detailedWordCount up)
• Have the librarian read more books (turn on more useXChunksForY toggles)
• Hire a smarter author (better summary model)

Each move improves quality, costs runtime.

---

Up Next

• Choosing Your AI Models — what runs the chunking + synthesis
• Customizing AI Prompts — tuning the actual instructions
• Performance & Slowness — when AI feels too slow