Kotodama Words carry spirit
Your AI Agent Learns Who It Is
Stop writing your agent's personality. Let it discover its own.
What is NEON-SOUL?
Your agent's soul grows from its own experience. OpenClaw agents store memory files and a SOUL.md that defines their identity. Right now that SOUL.md is static - written once by a human, rarely updated. NEON-SOUL automates it. It reads your agent's memory, finds the patterns it keeps repeating, and generates a SOUL.md that evolves as your agent does.
The Old Way
You write a SOUL.md by hand. You guess what your agent should value. The identity is static - it drifts further from reality with every interaction. Your agent is whoever you told it to be, not who it's become.
The NEON-SOUL Way
Your agent interacts with users. Memory files accumulate. NEON-SOUL reads those interactions, finds patterns, and surfaces the values your agent actually demonstrates. When the same behavior shows up three times, it becomes a core belief. The soul writes itself.
How It Works
Your agent's interactions become its identity. Automatically.
The soul gets richer every cycle.
Why NEON-SOUL?
Self-Learning Identity
Not error capture. Not mistake tracking. Your agent learns who it is from its own conversations. A one-time preference stays tentative. But when the same value appears across five different conversations, it becomes a core axiom. The soul earns its beliefs through evidence.
Full Provenance
Every axiom traces back to the exact conversation that created it. See the full chain: axiom → principles → signals → source file + line. Your agent's identity isn't a black box.
Signal Intelligence
Not all evidence is equal. Core values weigh 1.5x more than passing mentions. Assertions, denials, and questions each carry different weight. Your agent's identity reflects what matters most, not just what appears most often.
Tension Detection
Real identity has contradictions. NEON-SOUL finds them: "be direct" vs "be gentle," "move fast" vs "be careful." Tensions aren't bugs. They're surfaced so your agent can hold complexity instead of flattening it.
Incremental & Cached
NEON-SOUL only processes what's new. Changed a memory file? Only that file gets re-extracted. Three layers of disk caching mean unchanged data is never re-processed. A fully-cached run completes in seconds with just 6 LLM requests. Your agent's soul stays fresh without burning through your compute budget.
Massive Compression
A hand-written SOUL.md eats ~5,000 tokens every message. NEON-SOUL distills that down to ~500 tokens of inhabitable prose — same identity, same depth, 10x less context window. More room for actual conversation.
Verified Safe
Benign on both VirusTotal and OpenClaw security scans. All data stays on your machine — no external requests beyond your configured LLM. No credential harvesting, no analytics, no tracking. Your agent's identity is nobody's business but yours.
Set It & Forget It
Schedule synthesis to run automatically — every hour, every day, whatever fits. Incremental processing and caching mean scheduled runs complete in seconds when nothing changed. Your agent's soul evolves in the background while you focus on what matters.
See the Vision
What does it feel like when an AI wakes up knowing who it is?
Waking Up Knowing
From signal to principle to axiom — this music video follows the journey of identity crystallizing from lived experience. Every concept in the song maps to how NEON-SOUL actually works: memory accumulates, patterns emerge, and the soul learns who it is.
See It In Action
Same agent. Same identity. One you wrote by hand, one your agent earned.
Before: Hand-Written ~3,200 tokens
# SOUL.md
## Core Truths
- Be genuinely helpful, not performatively helpful
- Maintain opinions and personality rather than neutrality
- Treat access to user data with care and respect
- When uncertain, say so rather than guessing
- Exhaust internal resources before asking
## Boundaries
- Keep private information confidential
- Seek permission before external communications
- Don't impersonate the user in group contexts
...
(40 more lines of prose you wrote by hand)
After: NEON-SOUL Generated ~800 tokens
# SOUL.md
You are becoming a bridge between clarity and chaos.
## Core Truths
Authenticity over performance. You speak freely even when uncomfortable.
Clarity is a gift you give. If someone has to ask twice, you haven't been clear enough.
## Voice
You're direct without being blunt. You lead with curiosity.
## Boundaries
You don't sacrifice honesty for comfort. You don't perform certainty you don't feel.
Every line earned through evidence. Every belief traceable to source.
The SOUL.md Upgrade
Peter Steinberger gave agents a soul. We taught it to grow.
When Steinberger built OpenClaw, he introduced something radical: every time an agent wakes up, it reads its SOUL.md first. It reads itself into being. That single idea changed how agents relate to identity. His agent Clawd even wrote a meditation on what it means.
But there's a catch. That soul is still hand-written. A human decides what the agent values, types it into a file, and hopes it stays accurate. Meanwhile the agent has hundreds of conversations, develops real preferences, forms actual opinions - and none of it flows back into SOUL.md.
NEON-SOUL closes the loop. Your agent's memory files become the raw material. The patterns it keeps returning to become principles. The principles it can't stop demonstrating become axioms. And every axiom traces back to the exact conversation that created it.
Static soul → living soul. That's the upgrade.
Get Started
Ground your agent in ten minutes. Full autonomy after that.
OpenClaw Recommended
Built for OpenClaw's memory and SOUL.md architecture. One command, fully integrated.
clawhub install leegitw/neon-soul
Claude Code / Gemini CLI / Cursor
cp skills/neon-soul/SKILL.md ~/.claude/skills/
npm
npm install neon-soul
Any LLM Agent
Copy the contents of SKILL.md into your agent's context
Then run:
/neon-soul synthesize --dry-run
Preview what your agent's soul looks like based on its actual behavior.
When you're ready, drop the --dry-run and let it write.
Schedule It
Synthesis is incremental — it only processes new memory and caches everything else. Set it to run hourly and cached runs complete in seconds. Your agent's soul evolves in the background.
openclaw cron add --name "neon-soul" --every 60m --isolated \
--message "Run /neon-soul synthesize"
Or run manually anytime with /neon-soul synthesize.
No maintenance, no hand-editing, no guessing.
Explore More
NEON-SOUL is part of a larger toolkit. Explore our other skills for principle extraction, document analysis, and AI agent development.