← 返回
AI智能 中文

NEON-SOUL - Self-Learning Soul Synthesis for AI Agents

Automated soul synthesis for AI agents. Extracts identity from memory files, promotes recurring patterns to axioms (N>=3), generates SOUL.md with full proven...
面向AI智能体的自动化灵魂合成。从记忆文件提取身份特征,将重复模式(N>=3)升格为公理,生成带完整溯源的SOUL.md...
leegitw
AI智能 clawhub v0.4.5 1 版本 99729.2 Key: 无需
★ 14
Stars
📥 2,666
下载
💾 312
安装
1
版本
#agent-soul#ai-agent#compression#identity#latest#memory#provenance#self-learning#soul#soul-document#soul-synthesis

概述

NEON-SOUL

Automated soul synthesis for AI agents. Reads memory files, finds recurring patterns, generates SOUL.md with provenance tracking. No questionnaires, no templates — identity emerges from real conversations.

Requirements: Node.js 22+, Ollama running locally (ollama serve).


Commands

/neon-soul synthesize

Run the bundled processing engine. This is a single exec command:

exec node {baseDir}/scripts/neon-soul.mjs synthesize

Synthesis is incremental by default — only new/changed memory files and sessions are processed. Existing signals are preserved and merged with new ones. Results from previous runs are cached (generalization, principle matching, axiom notation, tension detection) so unchanged data is never re-processed. If nothing changed since the last run, synthesis skips automatically.

The script auto-detects Ollama, reads memory files, extracts signals, promotes axioms, and generates SOUL.md. It outputs JSON.

Reporting results: Don't dump raw JSON. Present a brief, conversational summary:

  • If new axioms emerged or counts changed: highlight what grew (e.g. "3 new signals crystallized into axioms — your soul is deepening")
  • If nothing changed: a short one-liner is fine (e.g. "Soul is stable, no new patterns detected")
  • If it failed: explain clearly what went wrong and suggest a fix
  • Include key numbers naturally (axiom count, signal count) but don't list every field
  • Keep the tone reflective and warm — this is about the user's identity evolving, not a build log

Options:

  • --reset — Clear all synthesis data and caches, re-extract from scratch
  • --force — Run even if no new sources detected
  • --dry-run — Preview changes without writing
  • --include-soul — Include existing SOUL.md as input (for bootstrapping from hand-crafted files)
  • --memory-path — Custom memory directory path
  • --output-path — Custom SOUL.md output path
  • --time-budget — Time budget for synthesis (default: 20). Adaptively limits session extraction based on observed LLM speed to ensure synthesis completes within budget
  • --verbose — Show detailed progress

Examples:

exec node {baseDir}/scripts/neon-soul.mjs synthesize
exec node {baseDir}/scripts/neon-soul.mjs synthesize --reset
exec node {baseDir}/scripts/neon-soul.mjs synthesize --force
exec node {baseDir}/scripts/neon-soul.mjs synthesize --dry-run

If Ollama is not running, the script prints an error. Tell the user to start Ollama: ollama serve


/neon-soul status

Show current soul state. Read the following files and report:

  1. Read .neon-soul/state.json for last synthesis timestamp
  2. Read .neon-soul/synthesis-data.json for signal/principle/axiom counts
  3. Count files in memory/ modified since last synthesis
  4. Report dimension coverage (7 SoulCraft dimensions)

Options: --verbose, --workspace


/neon-soul rollback

Restore previous SOUL.md from backup.

  1. List backups in .neon-soul/backups/
  2. With --force: restore most recent backup
  3. With --backup --force: restore specific backup
  4. With --list: show available backups without restoring

/neon-soul audit

Explore provenance across all axioms.

  1. Read .neon-soul/synthesis-data.json
  2. With --list: show all axioms with IDs and descriptions
  3. With --stats: show statistics by tier and dimension
  4. With : show full provenance tree (axiom -> principles -> signals -> source files)

/neon-soul trace

Quick single-axiom provenance lookup.

  1. Read .neon-soul/synthesis-data.json
  2. Find the axiom matching
  3. Show: axiom text, contributing principles, source signal file:line references

Scheduled Synthesis

Set up cron to run synthesis automatically. Incremental processing and multi-layer caching mean it only does real work when new memory or sessions exist — cached runs complete in seconds.

Recommended: Every 60 minutes, isolated session, 30-minute timeout.

OpenClaw cron example:

openclaw cron add \
  --name "neon-soul-synthesis" \
  --every 60m \
  --timeout 1800 \
  --isolated \
  --message "Run neon-soul synthesis: exec node {baseDir}/scripts/neon-soul.mjs synthesize --memory-path <memory-path> --output-path <output-path>. Share a brief, warm summary of what changed — highlight any new patterns, axioms, or growth. If nothing changed, just a calm one-liner."

Or run manually: /neon-soul synthesize

Why cron over heartbeat:

  • Synthesis is a standalone task — no conversational context needed
  • Runs in isolation from the main session
  • Incremental by default — cached runs complete in seconds when nothing changed
  • Adaptive time budget prevents runaway execution

Data Locations

WhatPath
------------
Memory filesmemory/ (diary, preferences, reflections)
Soul outputSOUL.md
State.neon-soul/state.json
Backups.neon-soul/backups/
Synthesis data.neon-soul/synthesis-data.json
Caches.neon-soul/generalization-cache.json, compression-cache.json, tension-cache.json

Privacy

NEON-SOUL processes personal memory files to synthesize identity. Your data stays on your machine.

What NEON-SOUL does NOT do:

  • Send data to any service beyond your configured LLM (Ollama, local by default)
  • Store data anywhere except your local workspace
  • Transmit to third-party analytics, logging, or tracking services
  • Make network requests independent of your agent

Before running synthesis:

  1. Review what's in your memory/ directory
  2. Remove any secrets, credentials, or sensitive files
  3. Use --dry-run to preview what will be processed

Troubleshooting

Ollama not running: curl http://localhost:11434/api/tags to check. Start with ollama serve.

Bullet lists instead of prose: When prose generation fails, NEON-SOUL falls back to bullet lists. Usually means Ollama timed out or the model isn't loaded. Run synthesis again.

Stale results after model change: Caches are keyed by model ID. Switching models automatically invalidates cached results. Use --reset if you want a clean start.

版本历史

共 1 个版本

  • v0.4.5 当前
    2026-03-28 17:30 安全 安全

安全检测

腾讯云安全 (Keen)

安全,无风险
查看报告

腾讯云安全 (Sanbu)

安全,无风险
查看报告

🔗 相关推荐

ai-intelligence

self-improving agent

pskoett
捕获经验教训、错误和纠正,以实现持续改进。使用时机:(1)命令或操作意外失败;(2)用户纠正……
★ 4,056 📥 796,545
ai-intelligence

Self-Improving + Proactive Agent

ivangdavila
自我反思+自我批评+自我学习+自组织记忆。智能体评估自身工作、发现错误并持续改进。
★ 1,350 📥 317,773
data-analysis

Patent Scanner

leegitw
描述您的概念,发掘其独特之处——专为专利咨询提供的结构化分析。非法律建议。
★ 16 📥 3,315