A self-evolution protocol and runtime for LLM-based agent systems.
A hierarchical multi-agent framework featuring RSPL and SEPL protocols for self-evolving agent loops.
Deep Research Agent is a self-evolution protocol and runtime built for LLM-based agent systems. It is designed for engineers and developers looking to implement structured multi-agent workflows with explicit resource and evolution layers.
The framework models prompts, agents, tools, environments, and memory as protocol-registered resources via the Resource Substrate Protocol Layer (RSPL). It uses a Self Evolution Protocol Layer (SEPL) to propose, assess, and commit improvements with auditable lineage and rollback capabilities.
It suits AI and systems developers building complex agentic workflows that require version tracking and self-improvement loops. It is not suited for users looking for a plug-and-play desktop assistant or those without experience configuring LLM environments.
Models prompts, agents, tools, environments, and memory as protocol-registered resources.
Specifies a closed-loop operator interface to propose, assess, and commit improvements.
MMEngine-style configurations to compose agents, tools, environments, and models consistently.
Records trajectories and manages iterative artifacts across runs for failure analysis.
The software is free under its MIT license, but you must supply your own LLM API keys and compute resources to run the agents and optimizers.
880 open remote AI-training roles, updated November 15, 2026; most pay $20–$175 an hour. Zekai lists them free; the application goes to the employer’s platform.
These reviewed AI tools do the same job instantly — from our scored directory.
RSPL models system components like prompts, agents, tools, environments, and memory as protocol-registered resources with explicit state, lifecycle, and versioned interfaces.
The SEPL layer provides a closed-loop operator interface to act, observe, optimize using reflection or RL methods, and remember insights into persistent memory.
See Zekai first in Google