Skip to content

bhilton114/Clair

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

34 Commits
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Clair — A Local Cognitive AI System for Real‑World Problem Solving A disciplined, modular cognitive architecture built on structured reasoning, verification, and honest uncertainty.

Clair is not a chatbot, not an LLM wrapper, and not a prompt‑engineering trick. It is a local cognitive system designed to solve real‑world problems through a structured, brain‑inspired reasoning pipeline with strict separation of responsibilities.

Clair’s core design principle is simple:

Never guess. Never hallucinate. Always verify.

🔍 Why Clair Exists Modern AI systems are fluent but unreliable. They blend perception, reasoning, memory, and validation into a single opaque process — which leads to hallucinations, false confidence, and unpredictable behavior.

Clair takes the opposite approach:

Every cognitive function is isolated.

Every stage has a defined role.

Uncertainty is detected and handled explicitly.

Verification is mandatory when confidence is low.

Memory is structured, traceable, and governed.

This creates a system that is:

transparent

predictable

self‑monitoring

resistant to hallucination

safe for real‑world tasks

🧠 Cognitive Pipeline Clair processes information through a strict, non‑overlapping pipeline:

Code Input → Perception → Affect → Reasoning → Calibration → Verification → Memory → Response Perception Extracts structure, intent, and problem type. No solving happens here.

Affect Assigns urgency, risk, and priority weighting.

Reasoning Generates candidate solutions using structured, multi‑step logic.

Calibration Evaluates uncertainty, detects conflict, and prevents false confidence.

Verification Validates claims using external checks, alternative reasoning paths, or internal consistency tests.

Memory Stores facts, outcomes, and confidence levels with provenance.

Response Produces the final answer only after all checks pass.

🔄 Three‑Loop Control System Clair is governed by three interacting loops:

  1. Reasoning Loop Iterative problem solving.

  2. Calibration Loop Uncertainty detection and self‑monitoring.

  3. Verification / Governance Loop Truth‑checking, conflict resolution, and safety.

This structure prevents self‑confirmation errors and hallucinations.

🧩 Key Features Local execution — no cloud dependency

Deterministic reasoning steps

Explicit uncertainty handling

Verification before output

Structured memory with confidence tracking

Modular architecture inspired by cognitive science

Honest “I don’t know” responses

No hallucination by design

🚀 Quick Start Install Code git clone https://github.com/bhilton114/Clair.git cd Clair Run an example Code python examples/solve_task.py More examples are available in the examples/ folder.

📚 Documentation Full documentation is available in the docs/ directory:

architecture.md

pipeline.md

memory-system.md

verification-loop.md

design-philosophy.md

🛣️ Roadmap See ROADMAP.md for upcoming features and long‑term plans.

🤝 Contributing Clair welcomes contributions that align with its philosophy of structured, honest, verifiable reasoning.

See CONTRIBUTING.md for guidelines.

🛡️ License Clair is released under the Apache 2.0 License, allowing broad use while protecting the integrity of the project.

Commercial licensing options will be available for enterprise use.

⭐ If you believe AI should be reliable, honest, and grounded — star the repo and follow the project.

About

A local cognitive AI system built on structured reasoning, verification, and disciplined memory — designed to solve real-world problems without hallucination.

Topics

Resources

License

Code of conduct

Contributing

Stars

Watchers

Forks

Releases

No releases published

Packages

 
 
 

Contributors

Languages