GBase
An agent framework that gives an AI agent long-term memory with active recall, a self-evolution cycle that rewrites its own prompts after failures, and multi-agent quality gates before code ships.
GBase is a framework for building agents that keep a long-term memory with active recall, rather than only the last few turns of a conversation, and distill each interaction into structured, reusable experience rather than raw logs.
Its recursive self-improvement cycle runs a stability audit and a performance evaluation, decides whether to roll back, and can rewrite the agent's own system prompt and tools between rounds. Tools are registered automatically from a Python decorator, quality gates route work through separate building, auditing and judging agents before it ships, an identity system swaps personas and tool sets per agent, and a built-in scheduler runs cron-like jobs without an external daemon.
A browser-based chat interface is included, started with a single command. It is written in Python.