Gizmo vs Clawdbot
Two personal AI assistants. Same underlying brain. Different approaches to memory and learning. Which one actually makes you more effective? Let's find out.
The Truth About AI "Intelligence"
You can't make an AI smarter. The intelligence comes from the model (Claude, GPT, etc.). What you can control is context—the information the AI has access to.
Both Gizmo and Clawdbot use Claude Opus 4.5. Same model, same reasoning capability, same knowledge. The difference isn't intelligence—it's how they manage what that intelligence can see.
What You Actually Control
Memory Systems
How past conversations and decisions are stored and retrieved
Learning Capture
How preferences and patterns are recorded over time
Bootstrap Context
What the AI knows about you before you say anything
Search & Retrieval
How efficiently the AI finds relevant past information
Memory Systems
Both systems use markdown files for memory. The difference is in structure and persistence.
Clawdbot
memory/ - Loads today + yesterday at session start
- Long-term memory only in private sessions
- Local storage only
- Auto-flush before context limit
Gizmo
sessions/ - Reads latest session + CLAUDE.md
- Incremental saves during session
- Git-backed, versioned, portable
- Auto-flush before context limit
Learning Capabilities
Neither system does neural learning—they don't update model weights. "Learning" means capturing information for future sessions.
Clawdbot Approach
- User prefers SvelteKit - Database password issue was URL encoding - Don't use cohort - Project uses Supabase - User is in Malaysia timezone
Gizmo Approach
## Preferences Discovered - User prefers SvelteKit (2026-01-10) ## Patterns That Work - URL encode special chars in DB passwords (2026-01-15) ## Avoid - Don't use "cohort" — use "intake" (2025-01-10)
Why This Matters
You forget to say "remember this." Gizmo captures learnings without being asked.
"Patterns That Work" and "Avoid" are different. Mixing them loses signal.
Knowing when you learned something helps judge if it's still relevant.
Memory Search
Having memory is useless if you can't find what you need. Both systems use hybrid search.
Vector Search (Semantic)
Finds conceptually similar content even with different words
BM25 Search (Keyword)
Finds exact matches for specific terms, IDs, error codes
Hybrid = Best of Both
Token Efficiency
Tokens cost money and fill up context windows. Efficient systems get more done with less.
Better context management means fewer tokens used while getting better results. You're not sacrificing intelligence—you're removing waste.
Where Tokens Go
Files loaded at session start. Gizmo's bootstrap is leaner.
With hybrid search, both find relevant sessions without reading everything.
Gizmo's structured preferences mean fewer "no, I meant..." corrections.
Net Token Savings
Final Verdict
The Bottom Line
For memory and learning: Gizmo wins. Automatic capture, structured categories, git-backed persistence, and better token efficiency make it the smarter system—even though the underlying AI is identical.
For messaging access: Clawdbot wins. If you need to chat with your AI via WhatsApp or Telegram, Clawdbot has that built in.
The insight: You can't change intelligence, but you can change everything around it. Better scaffolding makes the same brain more effective.