How StackMemory compares
Every AI coding tool loses context between sessions. Here's how the common approaches stack up.
| Capability | StackMemory | CLAUDE.md | .cursorrules | Vanilla context |
|---|---|---|---|---|
| Survives /clear | Yes | Yes | Yes | No |
| Survives session restart | Yes | Yes | Yes | No |
| Records decisions automatically | Yes | Manual | Manual | No |
| Scoped context (not flat) | Call stack | Flat file | Flat file | Linear |
| Full-text search | FTS5 + BM25 | No | No | No |
| Task management | Built-in + Linear | No | No | No |
| Multi-agent support | Team + Cord | No | No | No |
| Editor support | Claude, Codex, OpenCode | Claude Code only | Cursor only | Any |
| Setup effort | 2 commands | Create file | Create file | None |
| Scales with project | Database-backed | File grows | File grows | Resets |
The core problem
AI coding tools are stateless. Every session starts from zero. You re-explain architecture, re-state constraints, re-describe what you tried. CLAUDE.md and .cursorrules help by injecting static instructions, but they're flat files — no search, no scoping, no automatic capture.
StackMemory treats context as structured data. Decisions, constraints, and progress live in a SQLite database with full-text search. Context is organized as a call stack — nested frames that scope work like function calls. The active path is "hot," everything else stays queryable but quiet.
When to use what
Use CLAUDE.md for
Static project instructions, coding standards, and tool preferences that rarely change.
Use StackMemory for
Dynamic context — decisions made during work, task progress, architectural choices, things that evolve session to session.
Use both together
CLAUDE.md for the "what" (rules, standards). StackMemory for the "why" and "when" (decisions, progress, history).
StackMemory replaces
Manual context pasting, "reminder" prompts, re-explaining project state, lost decisions after /clear.