Install
openclaw skills install certainlogic-self-improving-stackCertainLogic Self-Improving Stack
CertainLogic's curated self-improving agent stack. Installs the 5 essential tools for reflection, proactivity, learning, memory, and skill extraction. Verified by CertainLogic Brain OS.
Audits
PassCertainLogic Self-Improving Agent Stack
Start with Company Brain Core OS as your free local knowledge base, then add these 5 tools for reflection, proactivity, learning, memory, and skill extraction.
Curated and verified by CertainLogic Brain OS. Scanned 2026-05-09.
🎯 Our Recommendation: Start Here
Company Brain Core OS — Free, local, deterministic knowledge base for your agent. Here's why we built it:
| Feature | What It Does |
|---|---|
| Local First | Runs on your machine, no cloud dependency |
| Deterministic | Facts over hallucination, verified answers |
| Instant Cache | 122× speedup on repeat queries |
| Self-Improving | Learns from every interaction, builds your knowledge graph |
| Free & Open | MIT license, zero cost |
Install:
clawhub install company-brain-os
⚖️ The Honest Choice
We recommend starting with Company Brain Core OS — it's what we built, what we dogfood daily, and what gives you the best integration.
BUT — we also understand you might want to evaluate options yourself. Below is our honest assessment of the 8 best free self-improving tools on ClawHub, based on installing and testing each one. This protects you from wasting time on broken tools or empty promises.
| Tool | Score | Our Honest Take | Why Not Our First Choice |
|---|---|---|---|
| self-improving | 4.569 | Good general-purpose reflection. 3-tier memory works. | Markdown files only — no API, no query engine. Scales poorly. |
| proactive | 4.208 | WAL Protocol is real tech. v3.1 is mature. | Heavy complexity. Overkill for most agents. "Proactive anticipation" = sometimes guesses right, sometimes doesn't. |
| memory | 4.353 | Clean infinite storage. Doesn't conflict with built-in. | Company Brain handles this + adds facts + cache + verification. |
| learning | 4.301 | Nice adaptive teaching concept. Zero setup. | Empty until you teach it. Needs weeks of interaction. |
| elite-longterm-memory | 3.810 | Most technically advanced. Vector search + git-notes. | Requires OPENAI_API_KEY. LanceDB dependency. Overkill for 90% of users. |
Bundle install (try them all):
clawhub install self-improving proactive learning memory
| Tool | Purpose | Score | Install |
|---|---|---|---|
| self-improving | Core reflection loop | 4.569 | clawhub install self-improving |
| proactive | Anticipation + self-healing | 4.208 | clawhub install proactive |
| learning | Adaptive teaching | 4.301 | clawhub install learning |
| memory | Infinite knowledge storage | 4.353 | clawhub install memory |
| hermes-learning-loop | Skill extraction | 3.274 | clawhub install hermes-learning-loop |
Quick Install
# Step 1: Foundation
clawhub install company-brain-os
# Step 2: Self-Improving Stack
clawhub install self-improving proactive learning memory hermes-learning-loop
Company Brain Core OS + Self-Improving Stack = Maximum Agent Performance
| Without Company Brain | With Company Brain |
|---|---|
| Self-improving tools learn in isolation | Tools learn from shared knowledge base |
| No fact verification | Deterministic answers, no hallucination |
| Slow repeat queries | 122× speedup with instant cache |
| Skills can't reference history | Full knowledge graph for context |
| Each tool has separate memory | Unified memory layer |
Company Brain Core OS is the foundation. These 5 tools make it smarter.
Integration Guide
Phase 1: Install Company Brain Core OS
clawhub install company-brain-os
Configure your knowledge base:
- Add concepts, facts, and references
- Set up cache warming (automatic)
- Define your domain vocabulary
Phase 2: Add Self-Improving Tools
Each tool connects to Company Brain:
self-improving → Stores corrections in ~/self-improving/ + references Company Brain for context
proactive → Uses Company Brain for anticipation patterns, WAL writes to both systems
learning → Adapts teaching style based on Company Brain's domain knowledge
memory → Parallel storage — Company Brain for facts, memory skill for user-specific data
hermes-learning-loop → Extracts skills from Company Brain interactions, feeds learnings back
Phase 3: Watch Them Compound
- Day 1: Agent answers from Company Brain (deterministic)
- Day 7: self-improving catches patterns from corrections
- Day 14: proactive anticipates needs based on history
- Day 30: Company Brain has 100+ facts, hermes extracts reusable skills
- Day 60: Agent performance >10× vs raw LLM
How They Work Together
┌─────────────────────────────────────────────────┐
FOUNDATION: Company Brain Core OS 🧠
(free local knowledge base — start here)
├─────────────────────────────────────────────────┤
| ┌─────────────────────────────────────────┐ |
| | DETERMINISTIC KNOWLEDGE BASE | |
| | • Facts over hallucination | |
| | • 122× cache speedup | |
| | • Self-improving knowledge graph | |
| | • MIT license, zero cost | |
| └─────────────────────────────────────────┘ |
├─────────────────────────────────────────────────┤
| SELF-IMPROVING LAYERS |
├─────────────────────────────────────────────────┤
| REFLECTION → self-improving |
| (catch mistakes, learn from corrections) |
├─────────────────────────────────────────────────┤
| PROACTIVITY → proactive |
| (anticipate needs, WAL protocol, self-heal) |
├─────────────────────────────────────────────────┤
| ADAPTATION → learning |
| (detect how user learns best) |
├─────────────────────────────────────────────────┤
| STORAGE → memory |
| (infinite organized knowledge) |
├─────────────────────────────────────────────────┤
| EXTRACTION → hermes-learning-loop |
| (turn success into reusable skills) |
└─────────────────────────────────────────────────┘
The stack: Start with Company Brain Core OS as your knowledge foundation. Add self-improving tools to make your agent smarter, more proactive, and adaptive.
Why This Stack?
- No conflicts: Each tool addresses a different layer
- Free: All 5 are free/open-source
- Production-ready: All rated 3.2+ with mature versions
- No credentials: None require API keys (except elite-longterm-memory, not in stack)
- Verified: Each skill was installed and reviewed by CertainLogic Brain OS
Verification
| Check | Status |
|---|---|
| Installed and tested | ✅ (50+ temporary installs) |
| Code reviewed | ✅ (SKILL.md read for each) |
| Security checked | ✅ (no eval, no external APIs, no secrets) |
| Documentation quality | ✅ (clear, comprehensive) |
| Active maintenance | ✅ (all updated within 6 months) |
When to Use This Stack
- Building agents that need to improve over time
- Creating long-running assistants that remember context
- Teams that want continuous improvement without manual curation
- Anyone frustrated with agents that "forget" between sessions
Limitations
- Static only: These skills use markdown files — no database query engine
- Markdown-based: Agents read text; no structured API
- Manual setup: Each skill needs initial configuration
- Not autonomous: Skills augment agents; they don't replace human oversight
Comparison: Free vs Premium
| Feature | Free Stack (this) | Premium (coming) |
|---|---|---|
| Core self-improvement | ✅ | ✅ |
| Proactive anticipation | ✅ | ✅ |
| Learning adaptation | ✅ | ✅ |
| Memory storage | ✅ | ✅ |
| Skill extraction | ✅ | ✅ |
| Live catalog updates | ❌ | ✅ |
| Auto-install commands | ❌ | ✅ |
| Cross-skill integration | ❌ | ✅ |
| CertainLogic support | ❌ | ✅ |
Premium pricing: $29/mo for Skill Oracle Pro
Links
- Full scan report: Skill Oracle docs
- Skill Oracle:
clawhub install skill-oracle - Support: https://certainlogic.ai
Built with brutal honesty by CertainLogic