Cognition

Rust implementation of a cognitively-grounded agent memory system, extended into a broader Agent architecture (not memory-only).

About This Repository

This repository is a Rust-first implementation inspired by CogMem design ideas, with a practical workspace structure for building:

• Long-term memory and retrieval
• Knowledge graph activation
• LLM-driven memory extraction
• Agent runtime components (skills, runtime orchestration, CLI, storage)

Reference to original CogMem source repository:

• https://github.com/triet4p/agent-memory-cognitive

Positioning

Cognition is not only a memory library.

In addition to cognitive memory components, the project includes Agent-oriented modules for:

• Runtime and orchestration
• Skill integration
• LLM provider abstraction
• CLI-driven operations and debugging
• Python binding pathway (planned via cognition-py)

Tech Stack

• Language: Rust (Edition 2024)
• Workspace: Cargo Workspace (multi-crate)
• Graph engine: petgraph
• Async runtime: tokio
• Storage: SQLite via sqlx
• Serialization: serde, serde_json
• Time and IDs: chrono, uuid

Workspace Structure

• crates/cognition-core: Shared contracts, types, model schemas, error/result abstractions
• crates/cognition-graph: Unified graph model and spreading activation logic
• crates/cognition-storage: Durable memory persistence (nodes, edges, migration)
• crates/cognition-memory: Memory extraction pipeline and prompt registry
• crates/cognition-llm: LLM integration layer (scaffold)
• crates/cognition-runtime: Agent runtime layer (scaffold)
• crates/cognition-skills: Skill subsystem (scaffold)
• crates/cognition-py: Python interoperability layer (scaffold)
• cognition-cli: Command-line entrypoint for local workflows

Current Cognitive Model Highlights

• 2-layer node schema for retrieval + lossless context:
  • narrative_fact (gist)
  • raw_snippet (verbatim)
• Intention lifecycle support:
  • status: planning, fulfilled, abandoned
  • deadline field for prospective memory
• Expanded memory network types:
  • world, experience, opinion, habit, intention, actioneffect
• Expanded edge types for cognitive transitions:
  • entity, temporal, semantic, causal, sr_link, ao_causal, transition
• SUM-based spreading activation with edge-type multiplier

Agent Capabilities Direction

Beyond memory, the project is structured to evolve into an end-to-end Agent system:

• Ingestion and extraction from conversational input
• Retrieval and activation-guided recall
• Runtime decision loop and tool/skill execution
• Multi-provider LLM integration
• Cross-language integration for Python ecosystems

Getting Started

Prerequisites

• Rust toolchain (stable)
• Cargo

Build

Run from repository root:

cargo check --workspace

Run root binary

cargo run

Run CLI binary

cargo run -p cognition-cli

Development Status

This codebase is under active development.

Some crates are production-oriented while others are scaffolded for upcoming Agent features. The workspace layout is intentionally modular to support iterative evolution from memory core to full Agent runtime.

Configuration

• Base config file: config/default.yaml
• Environment file: .env (local)

Documentation

Project documents are available in docs, including:

• CogMem.md
• cognition-core.md
• coginition-storage.md
• Project-Structure.md
• Project-Proposal.md

Contributing

Contributions are welcome.

Recommended flow:

1. Open an issue describing the change
2. Create a focused branch
3. Add tests for behavior changes
4. Submit a PR with clear rationale and impact notes

License

No license file is currently declared in this repository. Please add a LICENSE file before public distribution.