A modern C++ toolkit for building intelligent agents with sophisticated memory and planning capabilities.
Gloom provides a comprehensive set of tools for developing AI agents with human-like memory systems, strategic planning, and adaptive behavior. Built with performance and flexibility in mind, it offers both high-level abstractions and low-level control.
- Episodic Memory: Temporal storage and retrieval of experience-based memories
- Semantic Memory: Graph-based knowledge representation with relationship tracking
- Memory Store: Generic memory management with importance-based retention
- Agent framework for building autonomous systems
- Environment abstractions for agent interaction
- State management and action planning
- Strategic goal planning and execution
git clone https://github.com/yourusername/gloom.git cd gloom mkdir build && cd build cmake .. make
#include <gloom/core/agent.hpp> #include <gloom/memory/episodic.hpp> #include <gloom/memory/semantic.hpp>
int main() { using namespace gloom;
// Initialize agent with memory systems
Agent agent;
memory::EpisodicMemory episodic(100); // 100 episodes capacity
memory::SemanticMemory semantic(1000); // 1000 nodes capacity
// Create an episodic memory
auto episode_id = episodic.create_episode({
{"location", "lab"},
{"task", "experiment"}
});
// Add a memory to the episode
episodic.add_memory(episode_id, {
.content = "Observed unexpected behavior in test case 3",
.importance = 0.8
});
return 0;
}
#include <gloom/memory/semantic.hpp>
void semantic_example() { using namespace gloom::memory;
SemanticMemory semantic(1000);
// Create nodes representing concepts
auto cat_id = semantic.create_node("animal", {
{"type", "mammal"},
{"species", "cat"},
{"domesticated", "true"}
});
auto dog_id = semantic.create_node("animal", {
{"type", "mammal"},
{"species", "dog"},
{"domesticated", "true"}
});
// Create relationships between nodes
semantic.add_relationship(cat_id, dog_id, 0.7); // Strong relationship
// Query related concepts
auto related = semantic.get_related_nodes(cat_id, 0.5);
}
#include <gloom/memory/episodic.hpp>
void search_example() { using namespace gloom::memory;
EpisodicMemory memory(100);
// Search for memories with specific context
auto results = memory.search({
.context = {{"location", "lab"}},
.content = "unexpected",
.start_time = std::chrono::system_clock::now() - std::chrono::hours(24)
}, 10); // Limit to 10 results
for (const auto& [id, memories] : results) {
for (const auto& memory : memories) {
std::cout << "Memory: " << memory.content << "\n";
std::cout << "Importance: " << memory.importance << "\n";
}
}
}
#include <gloom/core/agent.hpp>
class CustomAgent : public gloom::Agent { public: void perceive(const Environment& env) override { // Process environmental input auto state = env.get_state();
// Store in episodic memory
auto episode_id = episodic_memory.create_episode({
{"state", state.to_string()},
{"timestamp", std::to_string(time(nullptr))}
});
}
Action decide() override {
// Query memories to inform decision
auto relevant_memories = episodic_memory.search({
.content = "success",
.limit = 5
});
// Use semantic memory for context
auto context = semantic_memory.search({
.concept = "current_goal"
});
// Make decision based on memories and context
return plan_action(relevant_memories, context);
}
private: memory::EpisodicMemory episodic_memory{1000}; memory::SemanticMemory semantic_memory{1000}; };
- C++17 or later
- CMake 3.15+
- Modern compiler (GCC 7+, Clang 6+, MSVC 2019+)
Comprehensive documentation is available in the docs/ directory:
- API Reference
- Architecture Guide
- Examples
- Best Practices
- Fork the repository
- Create a feature branch
- Commit your changes
- Push to the branch
- Create a Pull Request
This project is licensed under the MIT License - see the LICENSE file for details.
Project is in active development. API may change before v1.0 release.
For questions and support:
- GitHub Issues: Project Issues
- Email: team@askgloom.com