Blog
Engineering notes, model evals, and deep dives from the CowAgent team.
A Five-Layer Self-Evolution Mechanism for AI Agents
How an Agent keeps improving across long-running tasks in CowAgent: five layers from basic memory and knowledge maintenance, smart context summarization, post-session review, and dream-based consolidation, to source-code self-update.
DeepSeek V4 Agent Eval: Six End-to-End Scenarios on CowAgent
How deepseek-v4-flash behaves inside CowAgent's full agent loop across six end-to-end tasks: planning, coding, memory, browser, knowledge base, and very long documents.