In depth
Working memory is the agent's scratchpad. Different from raw context (which is just the conversation history), working memory is structured: it has a goal, a todo list, named facts, and survives compaction. On Digitorn working memory is a module, the agent calls remember/recall/forget/todo_add/todo_update like any other tool. Useful for multi-step tasks that need to remember what they decided five turns ago.
Related concepts
CompactionA runtime process that summarises old turns once the context window is nearly full, freeing space for new turns.Goal pinningA pattern where the original goal is auto-injected at the top of every turn so the LLM cannot lose the thread.Persistent memoryState that survives across sessions, scoped per user or per agent.
Newsletter
Get the next post in your inbox.
Engineering notes from the Digitorn team. No marketing, no launch announcements, no "10 prompts that will change your life". Just the things we write that we'd want to read.