If you use AI coding assistants, you probably repeat the same mistakes across projects.
PastChats Memory fixes that by giving your agent long-term memory:
- It indexes your old conversation history.
- It stores keyword index + vectors in SQLite.
- Before a new task, it recalls what worked and what failed.
Important: it does not auto-log your chats by itself. You index files (or automate indexing).
What “memory” looks like (example)
You run:
pastchats-memory recall --db .swarm/prompt_memory.db --query "webhook retry idempotent"
You get:
# Memory Recall
Query: webhook retry idempotent
## 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 designed to be pasted into your agent context before coding.
Why use it
- Fewer repeated mistakes
- Faster starts on new tasks
- Reuse proven patterns from old projects
- Keep memory local (SQLite file)
What it does under the hood
- Lexical search: FTS5 (fast keyword matching)
- Semantic search: vector similarity (sqlite-vec when available)
- Hybrid rank: combines both signals
- Recall mode: prints practical “what worked” context with sources
Start in 5 minutes
- Copy/paste the Vibe Coder Cheat Sheet
- Run the exact commands in Quick Start
- Use Search & Recall
- Read Auto-Capture if you want it to stay updated automatically
- Read How It Remembers (Simple)
- Read Workflow While You Work for the practical loop
- If you use Claude Code, set up Claude Code Hooks
Next: common patterns
If you like recipes, go straight to Common Workflows.
Who this is for
- Solo builders
- Teams using Claude Code / coding agents
- “Vibe coders” who want practical defaults and copy-paste commands