Neural Memory — Compact Context + Session Save

Get a token-budgeted (~500 token) snapshot of the current project's knowledge graph, and save a richer persistent snapshot for cross-session continuity.

What this does

Returns in one call:

• Staleness status — is the index current?
• Project overview — what this codebase does
• Active bugs — open, non-archived bugs
• Active tasks — pending/in-progress/testing, non-archived tasks
• Relevant nodes — code nodes semantically matching your query (if provided)

And saves to .neural-memory/session_context.md:

• Active tasks and bugs with their code node connections (file + line)
• Last 5 git commits
• Top 5 most important code nodes

When to use

• At the start of any task to orient yourself
• At the end of a session to persist context for the next session
• Between major steps to check active bugs and tasks
• Before modifying a specific area (pass that area as query_hint)

How to call

Via MCP tool (snapshot + save):

Tool: neural_save_context
{ "token_budget": 800 }

Via MCP tool (snapshot only):

Tool: neural_context
{ "query_hint": "authentication middleware", "token_budget": 500 }

Minimal call (no query hint — overview + bugs + tasks only):

Tool: neural_context
{}

Parameters

Session Persistence

The saved snapshot at .neural-memory/session_context.md is automatically:

• Written at session end (Stop hook) — no manual action needed
• Loaded at session start (UserPromptSubmit hook) when no query hint is present
• Refreshed whenever you run /neural-context manually

This gives new sessions immediate deep context without re-discovery.

Lifecycle

Bugs marked bug_status='fixed' and tasks marked task_status='done' are archived automatically at session end. Archived items:

• Do not appear in neural_context output
• Do not appear in neural_query by default
• Are still findable with neural_query using include_archived=true
• Can be manually archived/unarchived with neural_archive

This keeps active context lean as your project grows.