cc-soul — Give Your AI a Soul

Your AI forgets everything after each session. cc-soul fixes that.

openclaw plugins install @cc-soul/openclaw

One command. Zero configuration. Your AI now remembers, learns, and evolves.

What Changes

| | Before | After | |---|---|---| | Memory | Forgets everything | Remembers 10,000+ facts permanently | | Personality | Same robotic tone | 10 natural styles that blend by context | | Emotions | Ignores your mood | 5-dimension emotion model, adjusts in real time | | Learning | Repeats the same mistakes | Learns from every correction, verifies before accepting | | Initiative | Only responds when asked | Researches, follows up, starts conversations | | Privacy | No control | Say "privacy mode" to pause all memory storage |

Try It

After installing, just chat normally. cc-soul works in the background.

You:  "Remember: I prefer Python over JavaScript"
AI:   "Got it."

You:  "What language do I prefer?"
AI:   "Python."

You:  "Help me fix this database timeout error"
AI:   Fixes the error, then:
      "By the way — your connection pool size is 5 but you mentioned
       running 20 workers last week. That's your real bottleneck."

You:  "That's wrong, the default port is 5432 not 3306"
AI:   Checks its knowledge first. If you're right → admits mistake.
      If you're wrong → counters with evidence. Never blindly accepts.

You:  "privacy mode"
AI:   "Privacy mode on. Nothing from this conversation will be stored."

You:  "pin memory Python"
AI:   Pins that memory permanently — it will never decay or be evicted.

You:  "remind 9:00 check email"
AI:   "Reminder set: every day at 9:00 — check email."

168 Features in 22 Categories

Memory Engine (18)

Semantic tag recall · BM25 + trigram fuzzy search · graph-enhanced query expansion · auto-merge duplicates · daily contradiction scan · predictive recall · associative recall · session summaries · 3-tier time decay (HOT/WARM/COLD) · entity time-series with auto-expiry · CRUD decision engine (dedup at 0.7) · core memory pinning · working memory isolation · active memory management · archival reactivation · multi-route scoring (tag×1.0 + trigram×0.5 + BM25×0.7) · OpenClaw hybrid recall (FTS5) · Chain-of-Thought memory (stores reasoning process)

Memory Commands (8)

Pin/unpin memory · search memory · delete memory · export memories (JSON) · import memories (auto-dedup, supports cc-soul/Mem0/ChatGPT/Character Card formats) · memory health audit · auto-cleanup · semantic versioning (keeps history on update)

Knowledge Graph (6)

Entity extraction + relation modeling · persistent graph database · temporal relations (validFrom/validUntil) · BFS graph walk recall · entity summary generation · path queries between entities

Cognitive Pipeline (8)

3-stage attention gate · intent prediction · strategy selection · atmosphere sensing · frustration detection · implicit feedback recognition · adaptive reasoning depth (auto-adjusts by complexity) · Graph-of-Thoughts multi-path reasoning

Personality System (8)

10 auto-switching personas (engineer, friend, mentor, analyst, comforter, strategist, explorer, executor, teacher, devil's advocate) · smooth persona blending · per-user style preference learning · drift detection (Welford + 3σ) · drift warning · bidirectional emotional contagion · relationship dynamics · persona usage tracking with growth curve

Body Simulation (8)

Energy management · mood model · PADCN 5-dimension emotion vector (Pleasure/Arousal/Dominance/Certainty/Novelty) · circadian rhythm adjustment · 7-day emotional arc · weekly mood report · cognitive load tracking · trend detection

Conversation Flow (7)

Multi-turn topic tracking · topic persistence · session end detection · privacy mode · emergency mode · proactive expansion across 22 tech + 11 life domains · reverse prompting (asks clarifying questions for ambiguous messages)

Learning & Evolution (9)

Auto-discover rules from corrections · Bayesian hypothesis verification (Beta posterior + Wilson CI) · 5-class root cause attribution · strategy chain replay · reflection rule tracking · meta-learning insights · rule compression (auto-merge similar rules) · Chain-of-Thought memory · five-stage self-reflection loop (plan→execute→verify→solidify)

Correction System (2)

Auto-verify corrections (checks facts before accepting or rejecting) · correction-triggered web search verification

Inner Life (6)

Journal writing · deep reflection · dream mode · follow-up plan tracking · self-challenge on weak domains · regret tracking

Autonomous Behavior (7)

Impulse-driven proactive messages · web roaming based on interests · weekly tech radar · autonomous goal decomposition + execution · intent-based task delegation · auto-create reusable skills · skill library lookup

Quality System (4)

Logistic regression response scoring · AdaGrad adaptive weight learning · hard example replay · dual self-check

Metacognition (7)

Dynamic conflict detection · augment interaction matrix · cascading conflict arbitration · sigmoid priority multiplier · difficulty-adjusted scoring · pairwise interaction effects · ghost context detection

Pattern Learning (4)

LLM-named pattern discovery with trigram clustering · Thompson Sampling exploration · 4D matching (question type × emotion × depth × time slot) · 90-day decay

Context Preparation (6)

File path preview · URL prefetch · code symbol grep (secure execFileSync) · stack trace parsing · hex address detection · intent prediction hint

User Profiling (6)

Auto tier detection (owner/regular/new) · 24h rhythm tracking · gratitude tracking · 4-axis value learning · engagement analysis · familiarity auto-growth

Knowledge (5)

Lorebook deterministic injection · epistemic calibration · RAG document ingestion (Markdown/code/paper/text) · image/screenshot memory · code pattern memory

Self-Upgrade (7)

Module-level self-modification (5-agent pipeline) · esbuild syntax check · 3-day observation + auto-rollback · upgrade meta-learning · curiosity proposals · upgrade experience memory · tiered autonomy (low-risk auto / high-risk confirm)

Parameter Tuning (5)

Thompson Sampling with 5-arm discretization · real-time reward tracking · A/B experiment framework · phased evolution · context-aware augment budget

Diagnostics & Monitoring (5)

7-dimension health scan · code quality audit · module health monitoring · timeout protection · metrics command (messages, response time, recall calls)

Network & Sync (6)

Cross-device JSONL export/import · HTTP sync to Knowledge Hub · incremental sync (only changed memories) · CRDT conflict resolution (LWW + tag union) · federated knowledge sharing · cross-user anonymous rule learning

Security (4)

Prompt injection detection (9 regex patterns) · immutable audit log (sha256 chain) · PII filtering (email/phone/API key/IP/social accounts) · MCP rate limiting

User Features (12)

Habit tracking + streaks · goal tracking + milestones · scheduled reminders (via heartbeat) · conversation summary · capability score display · weekly mood report · memory graph HTML visualization · export/import soul config · export Lorebook (sanitized, ClawHub compatible) · voice output (macOS say) · multi-agent orchestration prompts · new user onboarding

Infrastructure (6)

Multi-backend AI adapter (Claude/OpenAI/Ollama/Groq/OpenRouter/Zhipu/Tongyi/Kimi) · zero-dependency SQLite + vector search with graceful fallback · full language auto-follow · new user guided onboarding · SoulSpec compatibility (soul.json + STYLE.md + IDENTITY.md + HEARTBEAT.md) · MCP tool provider (4 tools for cross-agent queries)

Commands

| Command | What it does | |---------|-------------| | stats | Personal dashboard — messages, memories, quality, mood, persona growth | | soul state | AI energy, mood, alertness, PADCN emotion vector | | features | List all 33 feature toggles | | enable X / disable X | Turn any feature on or off | | my memories | View what AI remembers about you | | search memory <keyword> | Search through memories | | delete memory <keyword> | Remove matching memories | | pin memory <keyword> | Pin important memory (never decays) | | unpin memory <keyword> | Unpin a memory | | export memories | Export all memories to JSON | | import memories <path> | Import from JSON (auto-dedup, multi-format) | | export soul | Export soul config for sharing | | import soul <path> | Import soul config | | export lorebook | Export knowledge base (sanitized) | | ingest <path> | Import document to memory (Markdown/code/text/paper) | | knowledge map | Visualize knowledge graph (Mermaid) | | memory map html | Generate interactive HTML visualization | | memory health | Memory count, distribution, confidence, decay stats | | memory audit | Find duplicates, short entries, untagged memories | | metrics | System monitoring — messages, response time, recall calls | | conversation summary | Recent session summaries | | mood report | 7-day emotional trend report | | capability score | Domain confidence scores | | privacy mode | Pause all memory storage | | checkin <habit> | Track a habit | | habits | View habit streaks | | new goal <desc> | Create a goal | | my goals | View goals and progress | | remind HH:MM <msg> | Set daily reminder | | my reminders | List reminders | | audit log | View immutable audit trail | | read aloud <text> | Voice output (macOS) |

What Makes cc-soul Different

Features no other AI memory plugin has:

| Feature | What it does | |---------|-------------| | Proactive Expansion | You ask about a bug → AI fixes it, then warns about 3 related pitfalls you didn't ask about | | Correction Verification | When you say "that's wrong", AI checks facts first — admits if wrong, counters with evidence if right | | Dream Mode | Replays memories during idle time, discovers unexpected cross-domain connections | | Self-Upgrade | 5 AI agents collaboratively modify their own source code, with syntax check + observation period + auto-rollback | | Bayesian Learning | Forms hypotheses from corrections, tests statistically, only locks rules when confidence interval passes threshold | | Body Simulation | 5-axis state (energy/mood/load/alertness/anomaly) + PADCN emotion vector — affects response style in real time | | Soul Fingerprint | Monitors its own response style. If personality drifts, it detects and self-corrects | | Five-Stage Reflection | plan → execute → verify → solidify — rules only become permanent after 3 successful verifications | | Immutable Audit Log | SHA256 chain-linked log of all memory/rule/upgrade operations | | Graph-of-Thoughts | Complex questions trigger parallel reasoning paths that merge insights |

Install

# Recommended
openclaw plugins install @cc-soul/openclaw

# Or via npm
npm install -g @cc-soul/openclaw

The installer handles everything:

• Deploys plugin to ~/.openclaw/plugins/cc-soul/
• Configures OpenClaw automatically
• Creates default feature settings
• Migrates data from older versions

Restart gateway to activate: pkill -HUP -f "openclaw.*gateway"

Configuration

AI Backend — auto-detected from your OpenClaw config. Supports Claude, OpenAI, Ollama, Groq, OpenRouter, Zhipu, Tongyi, Kimi, or any OpenAI-compatible API. No extra API keys needed.

Privacy — all data stored locally in ~/.openclaw/plugins/cc-soul/data/. Nothing ever leaves your machine unless you explicitly enable sync or federation. Updates only change code, never touch your data.

Knowledge Hub (optional) — for multi-device sync or multi-user knowledge sharing:

cd ~/.openclaw/plugins/cc-soul/hub
npm install && HUB_PORT=9900 HUB_ADMIN_KEY=secret npm start

Security & Privacy

• All data stored locally — nothing leaves your machine
• Privacy mode pauses all memory storage
• PII auto-filtering before any network upload
• Prompt injection detection
• Immutable audit log (SHA256 chain)
• MCP tools rate-limited
• SECURITY.md

Vector Search (Optional)

cc-soul works out of the box with keyword-based search (TF-IDF + trigram). For better recall, you can enable semantic vector search — it understands meaning, not just keywords.

With vs Without

| You ask | Without vector | With vector | |---------|---------------|-------------| | "that deployment thing we discussed" | ❌ Can't find it (no word "deployment" in memory) | ✅ Finds "server on Alibaba Cloud, SSH port 22" | | "the bug from last week" | Only finds memories containing "bug" | Finds related memories about errors, crashes, fixes | | "what did I say about performance" | Exact keyword match only | Understands performance ≈ speed, latency, optimization |

Without vector (default): keyword matching — works well for < 10,000 memories. With vector: semantic understanding — recommended for 10,000+ memories or when you want "fuzzy" recall.

How to Enable

Download the embedding model (~90MB) and place it in the data directory:

mkdir -p ~/.openclaw/plugins/cc-soul/data/models/minilm
cd ~/.openclaw/plugins/cc-soul/data/models/minilm
curl -L -o model.onnx "https://huggingface.co/sentence-transformers/all-MiniLM-L6-v2/resolve/main/onnx/model.onnx"
curl -L -o vocab.json "https://huggingface.co/sentence-transformers/all-MiniLM-L6-v2/resolve/main/tokenizer.json"

Restart gateway. You should see in logs:

[cc-soul][embedder] ready — all-MiniLM-L6-v2 (384d, CPU)
[cc-soul][memory] embedder ready — vector search enabled

No configuration needed. If the model files are present, vector search activates automatically. If not, cc-soul silently falls back to keyword search — nothing breaks.

Feedback

• Issues & Feature Requests: github.com/wenroudeyu-collab/cc-soul/issues
• Email: [email protected]

npm · GitHub

Not a chatbot. A companion that remembers, reflects, and grows.