KlomboAGI: An Autonomous Cognitive Runtime for Persistent Agent Research

KlomboAGI is a Python-based experimental autonomous cognitive runtime system designed specifically for persistent agent research. It does not claim to implement AGI; instead, it explores the possibility of making agents smarter over time through mechanisms like persistent memory, world models, and a planning-validation-criticism loop. The core research question is: Can agents improve task performance through long-term operation and continuous learning? The project is positioned as a serious research platform, emphasizing the value of persistence and the time dimension.

Core Objectives

Focus on the long-term accumulated value of agents, distinguishing from one-time intelligent systems, assuming their value lies in long-term experience accumulation and behavior optimization.

Comparison with Similar Projects

• AutoGPT: More emphasis on persistence and structure, not exploratory
• LangChain: Focuses on long-running agents, not immediate response
• BabyAGI: Provides a more complete execution environment and world model

The project clearly states 'not AGI' and positions itself as a research platform rather than a product with an honest attitude.

Persistent Storage

• Task storage: Persistence of missions, tasks, world states, etc.
• Multi-layer memory: Working/semantic/procedural memory (drawing on cognitive science)
• World model: Maintains entities, relationships, and snapshot history

Cognitive Cycle Architecture

Planning-Validator-Critic loop:

• Planner: Decomposes goals into steps
• Validator: Checks safety and compliance
• Critic: Reflects, evaluates, and improves

Other Features

• Protected execution: Trace analysis + policy checks
• Workspace operations: Interaction with files/commands/code repositories
• Scheduling queue: Priority management + CLI support

Technical Implementation

Python3.9+, BSL license, file system priority, modular architecture, test-driven.

Test suite coverage:

• Runtime initialization and persistence
• Mission/task tracking
• Cognitive components (memory/planning/criticism, etc.)
• World model updates
• Safety policies
• Real operation execution
• Multi-step loops and stop conditions
• Code repository evaluation

Testing ensures the project is verifiable and trustworthy, distinguishing it from experimental code.

LLM Integration

• Optional integration for intelligent planning/criticism/reflection
• Compatible with OpenAI-style APIs (Ollama/OpenAI/Groq, etc.)
• No external dependencies; HTTP calls use standard libraries

Graceful Degradation

Automatically switches to heuristic rules when LLM is unavailable to ensure robustness.

Quick Start

1. Configure the .env file
2. Run diagnostics: klomboagi doctor
3. Initialize: klomboagi init
4. Create and run a mission: mission create + run

Application scenarios:

• Long-term code analysis (monitoring code repository changes)
• Automated document maintenance (generating change summaries)
• Agent behavior research platform
• Personal knowledge management assistant

It is recommended that relevant developers/researchers use this platform for experiments.

KlomboAGI represents the direction of agent research shifting from immediate response to long-term accumulation, with solid design and comprehensive testing. Its core value lies in raising the question of 'how to make agents grow' and providing an experimental platform for related fields.