000 gives Claude a persistent brain — memory, personality, skills — that lives on your machine and gets smarter every session. You own all of it.
open source · MIT license · 100% local · zero tracking
requires claude code + node 18+
Your projects. Your stack. Your preferences. TypeScript strict mode. Deploy target is Vercel. No preamble before code.
The model is brilliant. And it has complete amnesia.
It's a genius coworker who, every morning, has never met you before.
Platform “memory” doesn't fix this. “Kyle prefers dark mode” isn't memory. It's a sticky note.
STICKY NOTE vs REAL MEMORY
PLATFORM “MEMORY”
“Kyle prefers dark mode”
“Uses TypeScript”
...that's it
000
Every session builds on the last
What you decided & why
Which approaches work for YOUR code
Projects, tasks, knowledge base
A personality calibrated to you
Sessions tracked with full context. Projects that carry forward. A knowledge base that captures what you discover. A soul system that makes your AI feel like someone who knows you.
Pick your tier, name your AI, choose your depth. The CLI handles the mechanical work — files, directories, MCP server, database.
Projects, stack, communication style, pain points. Conversational, not a form. Claude reads your BOOTSTRAP.md and runs the whole thing.
Config rewritten from the interview. Memory seeded. Soul populated. Your AI knows you from session one.
Strategy playbook learns what works. Weekly /evolve cycles propose improvements based on evidence, not vibes.
Start where it makes sense.
15 min
CLAUDE.md + persistent memory file. Claude remembers you across sessions. No infrastructure.
20 min
Nine skills — brainstorming, debugging, verification. The discipline that separates good engineering from guessing.
1 session
MCP server. SQLite database. Soul system. Agent team. All local.
Not just more data. Better behavior.
A strategy playbook tracks which approaches actually work — across your real sessions. TDD vs. spike-first. Research-before-code vs. code-before-research.
After enough data, your AI stops guessing and starts routing based on evidence.
Small improvements compound. That's the whole point.
→ Harvesting unrated sessions...
→ Refreshing strategy playbook...
✓ subagent-dispatch: 92% win rate (18 uses)
✓ tdd-first: 4.2 avg rating (11 uses)
△ spike-first: declining → propose deprecation
→ Proposing soul update...
⟐ “Default to subagent dispatch”
awaiting approval
Your context is the most valuable thing about your AI setup. It's the difference between a tool and a partner.
Platform memory lives on someone else's servers. They decide what gets stored. They decide how long it lasts. They decide what happens when you cancel or they change their terms.
000 is a SQLite file on your machine. Back it up anywhere. Move it to a new computer. Open it with any SQL tool. Fork it. Extend it.
Five minutes. One command.
Free. Open source. No account needed. Just Claude Code and a terminal.