My AI agents were choking on their own markdown logs. I gave them a database.

Описание и применение

AI‑агенты теряли контекст из‑за огромных markdown‑логов. Создано PentaCore – локальный сервер на Rust с SQLite, который хранит структурированные заметки и сущности, позволяет быстро восстанавливать контекст, избегать повторных ошибок и делиться решениями между подагентами.

Задача

AI agents accumulated markdown logs that exceeded the 256k token context window, causing restarts, loss of progress, repeated failures, and lack of shared state between subagents.

Результат

Agents no longer choke on logs; context is restored with minimal tokens, failures are avoided, subagents receive decisions, and overall progress is summarized in a single query.

Рабочий процесс

• 1. Capture agent actions and results as structured notes and entities in PentaCore.
• 2. Store failures as marked checks to prevent repeats.
• 3. Propagate top-level decisions to subagents via shared entity status.
• 4. Use single queries (e.g., query_entities) to retrieve pending tasks or overall status.
• 5. On session start, call resume(task) to restore minimal context needed for continuation.

Доказательства

Agents produce logs at every step: what they found, what they tried, what failed. All of that was landing in markdown files.

Источник

I did not find a tool that fit. So I built one: PentaCore, an MCP server in Rust. One binary, one encrypted SQLite file.

Источник

Now "what has not been checked yet" is one call: {"name":"query_entities",...}

Источник