What is cognitive-engine?

cognitive-engine is a TypeScript library that gives AI agents actual cognitive capabilities - not just prompt engineering.

The Problem

Most AI agent frameworks are wrappers around LLM API calls. They have no memory between sessions, no ability to learn from mistakes, and no self-awareness.

// This is what most "agents" look like
const response = await llm.complete([
  { role: 'system', content: prompt },
  { role: 'user', content: input }
])

The Solution

cognitive-engine provides a full cognitive pipeline:

• Perception - understands context, not just text
• Episodic Memory - remembers what happened (conversations, outcomes, mistakes)
• Semantic Memory - stores what it knows (facts, patterns, domain knowledge)
• Working Memory - keeps relevant context active
• Reasoning - forms beliefs from evidence, generates intentions
• Metacognition - monitors its own thinking, catches biases

Architecture

Input -> Perception -> Working Memory -> Reasoning -> Response
              |              ^
              v              |
         Episodic        Semantic
         Memory          Memory

Each module is a separate npm package. Use all of them together via the orchestrator, or pick individual modules.

Key Design Decisions

• Provider agnostic - any LLM, any storage. Implement LlmProvider and Store interfaces.
• Zero casts - strict TypeScript with no type assertions anywhere in the codebase.
• Modular - use the full pipeline or just the parts you need.
• Tested - 320 tests across 23 test files.