How It Remembers (Simple)

This tool does not “magically learn your brain”. It does something simpler and more reliable:

You point it at chat history files.
It saves those conversations into a local database.
Later, you ask a question and it finds the most similar past moments.
It prints those moments in a readable way before you start coding.

What gets captured

PastChats Memory captures the text of your past conversations as turns:

user: what you asked
assistant: what the AI answered

It also keeps basic metadata so results are trustworthy:

which file it came from
which “project” it came from (inferred from folder)
which turn number inside the conversation

You control what it captures by choosing what to index:

pastchats-memory index --db .swarm/prompt_memory.db --input ~/projects

If it’s not in the indexed folders, it won’t be remembered.

Does it auto-log my chats?

Not by itself.

PastChats Memory does not intercept chat messages. It only indexes files you point it at.

To make it automatic, set up scheduled indexing (see: Auto-Capture).

How it becomes searchable

It builds two ways to search the same text:

Keyword search (fast, exact-ish)
“Meaning” search (find similar ideas even if the words differ)

Then it combines both into one final ranking.

You don’t need to understand the math. You just write better queries:

<problem> + <constraint> + <stack>

Example:

webhook retry idempotent python

Why recall output is readable

recall is designed to be copy-paste context for an agent.

When it finds a past user prompt that looks relevant, it also grabs the next assistant reply and labels it:

Prompt: (what you asked back then)
What worked: (the answer you got back then)
Source: (where it came from)

Example output

# Memory Recall
Query: retry strategy

## Memory 1 [0.190] - my_project
Prompt: build retries with idempotency keys
What worked: use bounded exponential backoff and store idempotency key state
Source: /path/to/chat.md (turn 0)

This is the “long-term memory” effect: it keeps you from re-learning the same lesson repeatedly.

What it does not do

It does not update an LLM’s training.
It does not guarantee the retrieved answer is correct.
It does not auto-capture new chats unless you index them.

Think of it like: “grep + semantic search across your past AI work, with clean output.”