Persistent Memory
SWUST Code's memory system lets your AI coding assistant remember project knowledge across sessions.
How It Works
Memory is stored as Markdown files and reconciled into a SQLite FTS5 full-text index. The core v2 runner loads memory context during system prompt assembly and injects global MEMORY.md content with a 4KB cap.
Memory Directory Structure
~/.local/share/swust-code/memory/
global/
MEMORY.md # Cross-project preferences
projects/
<project_hash>/
MEMORY.md # Project knowledge
facts/
<fact>.md # One-fact-per-file store
sessions/
<session_id>/
checkpoint.md # Session checkpoint (11 sections)
notes.md # Temporary notesMemory Tools
memory— Search persistent knowledge in the core v2 registrymemory_write— Write structured knowledge to memory files in the core v2 registry
Automatic Behavior
- Pre-search sync: Automatically syncs disk files to FTS index before each search
- Context injection: Auto-injects MEMORY.md content into system prompt (4KB cap)
- Incremental sync: Fingerprints based on file size and mtime, only processes changed files
- Imports:
MEMORY.mdsupports@pathimports during local resolution