Player not loading? Watch on YouTube
This Portuguese-language advanced Hermes Agent tutorial covers separate profiles, external memory and task orchestration. The presenter creates a student profile in the dashboard, then completes its command-line setup after reporting problems with profiles that appeared active but failed to load. The demonstration uses Ubuntu alongside Windows and PowerShell. Each profile has separate configuration, sessions, memories and skills; the presenter says each supports one external memory provider.
OpenViking is described as a context layer that organizes resources and retrieves relevant material through semantic search. Hindsight adds entity relationships and a knowledge graph, with retain, recall and reflect operations. The lesson configures lightweight CPU embeddings for OpenViking and shows Hindsight Local Embedded. After a CUDA compatibility error, the presenter forces Hindsight embeddings and reranking onto the CPU. Retrieval becomes slower, and an initial recall misses a newly stored memory before a later attempt succeeds after indexing.
The setup combines local memory processing with hosted model access because the presenter's machine cannot run Ollama while recording. It is not an entirely offline demonstration. Later sections connect Codex as a coding assistant inside Hermes, attempt subagent delegation, test communication between profiles through A2A, and create a Kanban task. The final example requests a daily AI news job for Telegram; delivery is not shown.
