Context Engineering for Claude Code
Persistent Memory for Claude Code
Your agent forgets everything between sessions. AXME Code gives it a structured, accumulating knowledge base that grows smarter with every session.
The problem: total amnesia
Every Claude Code session starts from zero. No matter how many times you've explained your project, the next session knows nothing.
Agents forget everything
Session 12: you pick Zod for validation. Session 13: the agent generates Joi schemas. Session 14: Joi again. There is no memory between sessions. Every conversation is a blank slate.
CLAUDE.md doesn't scale
10 lines? Works great. 200 lines? Claude ignores 80% of it. There's no structure, no priority, no enforcement. It's a static text file that you have to maintain by hand.
Same mistakes repeat
You correct the agent: "don't use sync HTTP in async handlers." Next session, it writes sync HTTP again. Feedback is never captured, never persisted, never applied.
No session continuity
Session 31 ended mid-refactor. Session 32 starts from scratch. You spend 15 minutes re-explaining where things stand before any work begins.
What AXME Code provides
Four systems that give your agent persistent, structured, accumulating memory. No manual maintenance required.
Oracle
Auto-scanned at setup. Your tech stack, project structure, naming patterns, API conventions, glossary of domain terms. The agent reads this once at session start and understands your project.
Memory
Feedback from your corrections and validated patterns that worked. When you say "don't do X, do Y instead," the agent saves it. Next session, it knows. The session after that, it still knows.
Handoff
When a session ends, the agent writes a structured handoff: what was done, what's in progress, what's next, which PRs are open. The next session picks up exactly where you left off.
Background Auditor
Close the window without saving? The auditor reviews the full transcript after the session ends. It extracts missed memories, decisions, and safety rules automatically.
How it differs from CLAUDE.md
CLAUDE.md is a starting point. AXME Code is a knowledge system.
CLAUDE.md
- ✕ Static — you write it, you maintain it
- ✕ Manual — nothing is extracted automatically
- ✕ No enforcement — all instructions are equal weight
- ✕ No memory from corrections — feedback is lost
- ✕ No session continuity — no handoff, no state
- ✕ Degrades at scale — 200+ lines get ignored
AXME Code
- ✓ Structured — oracle, memories, decisions, handoff
- ✓ Automatic — scanned at setup, extracted during work
- ✓ Enforcement levels — required or advisory
- ✓ Learns from corrections — mistakes saved as memories
- ✓ Session continuity — handoff picks up where you left off
- ✓ Scales with usage — knowledge accumulates, never lost
Comparison
How AXME Code stacks up against the alternatives.
| Feature | CLAUDE.md | claude-mem | AXME Code |
|---|---|---|---|
| Auto-scanned project context | ✕ | ✕ | ✓ Oracle |
| Persistent memories from corrections | ✕ | ✓ | ✓ |
| Structured decisions with enforcement | ✕ | ✕ | ✓ Required / Advisory |
| Safety hooks (hard enforcement) | ✕ | ✕ | ✓ |
| Session handoff | ✕ | ✕ | ✓ |
| Background audit (missed items) | ✕ | ✕ | ✓ |
| Multi-repo workspace support | ~ Manual | ✕ | ✓ |
| Scales beyond 200 lines | ✕ | ✓ | ✓ |
| Deduplication | ✕ | ✕ | ✓ |
Give your agent a memory
One command to install. One command to setup. Your agent never forgets again.
# Install
curl -fsSL https://raw.githubusercontent.com/AxmeAI/axme-code/main/install.sh | bash
# Setup your project
cd your-project && axme-code setup
# Use Claude Code as usual
claude