Deep dives, practical guides, and incident analyses from engineers who build Safeguard. No fluff, no vendor FUD — just what you need to ship secure software.
DGM, ADAS, Gödel Agent and SICA let an LLM agent rewrite its own code and keep changes that score better. What each changed, cost and risked.
Continual learning means a deployed model or agent keeps changing. How feedback loops, forgetting and drift break it, and how to monitor and roll it back.
Nous Research's Hermes models run from a 2023 LLaMA 13B fine-tune to Hermes 4.3 on Seed-OSS-36B. Every version, base, licence and how to inventory one.
Eight ways agents improve themselves, from reflection to weight training: what each one changes, who checks the change and how you roll it back safely.
Record model, agent version, skills, task, tools and human approver for every agent-made commit, skill and memory change, then sign it as SLSA provenance.
Published systems improve their own prompts, skills and agent code on fixed benchmarks. None shows a compounding loop that retrains its own model.
Self-improving agents learn by rewriting notes, memory and skill code between tasks, not weights. How Reflexion, Voyager and MemGPT work, and the risks.
A self-extending agent installs packages and writes skills, scripts, cron jobs and MCP configs. Treat each as a supply chain artifact: pin, review, sign.
Hermes Agent learns through two capped memory files, self-written skills, SQLite session search and a background review. How each works and is gated.
Weekly insights on software supply chain security, delivered to your inbox.
Your first fix PR is minutes away.
No sales call required, even your agent can complete the purchase over MCP.
