Memory
Nenjo's scoped memory system — core, project, and shared focus — plus artifacts for saved outputs that persist across runs.
Memory is learned context that persists across runs. Where knowledge graphs hold explicit source material and context blocks hold static guidance, memory captures what an agent has learned: preferences, corrections, decisions, and recurring patterns.
The three scopes
| Scope | What it stores | Typical use |
|---|---|---|
core_focus | Long-term, cross-project knowledge | Personality, methodology, core principles |
project_focus | Context specific to one project | Task history, decisions, feedback |
shared_focus | Knowledge shared across projects | Reusable patterns, team conventions |
An agent's memory profile selects which scopes it draws from. The profile is mutable retrieval and memory-writing configuration, not model-visible prompt content.
Memory in model context
The worker retrieves in-scope memories when a session context epoch begins and stores the resulting snapshot in the session-context message. Replayed turns reuse those exact bytes. A changed profile affects a new session epoch instead of rewriting the prefix of an existing conversation.
Memory and memory-profile selectors are therefore not available inside system prompts, developer prompts, or context blocks.
Artifacts
Artifacts are saved generated outputs or workspace deliverables — distinct from memory, knowledge graphs, and package source material. Where memory is learned context, an artifact is a produced output an agent keeps so later work can see it.
Artifacts arrive as typed chat/task inputs or are inspected with artifact tools. They are not template variables. Use them when later work needs a saved output or deliverable, but do not treat artifacts as source-of-truth knowledge or as learned memory.
Best practice
Start simple with only project_focus. Expand to core_focus and shared_focus as an agent's responsibilities grow across projects.
Keep the boundaries clear:
- Memory — learned, evolving context.
- Knowledge graphs — explicit source material to search and cite.
- Context blocks — static reusable guidance.
- Artifacts — saved generated outputs and workspace files.