palinode-memory

Palinode Memory

Palinode stores typed memory as git-versioned markdown files with hybrid search (BM25 + vector). This skill provides procedural guidance for using Palinode's tools effectively.

Core Tools

Tool	When to Use
palinode_search	Find relevant memories by meaning or keyword
palinode_save	Store a new typed memory (person, project, decision, insight)
palinode_status	Show file counts, chunk counts, entity graph, system health
palinode_trigger	Register an intention — auto-surfaces a memory file when context matches
palinode_diff	See what changed in a memory file recently
palinode_consolidate	Run or check the weekly compaction job

Memory Types

Use the right type when saving:

• PersonMemory — facts about a person (name, role, preferences, context)
• Decision — architectural/product decisions with rationale
• ProjectSnapshot — current state of a project (milestone, blockers, next)
• Insight — generalizable lessons learned (applies beyond current project)
• ActionItem — follow-up task with owner and deadline

Search Effectively

Palinode uses 4-phase injection — you don't always need to search manually:

1. Core files (core: true) are always injected at session start
2. Topic search runs on every turn automatically
3. Associative recall expands entity mentions automatically
4. Triggers fire when registered contexts match

Only call palinode_search explicitly when you need something specific not surfaced by injection, or when the user asks "what do I know about X?"

Saving Memories

Guidelines:

• Save decisions with rationale — not just what was decided, but why
• Save people context when you learn preferences, constraints, or relationships
• Save insights when a lesson applies broadly (not just to one task)
• Don't save raw conversation transcript — save the distilled fact

Example save calls:

palinode_save(
  content="Alice and Bob are co-founders, 3-year gap in joining. CEO role is shared.",
  type="PersonMemory",
  entities=["person/alice", "project/my-app"]
)

palinode_save(
  content="Curation > volume for training data. 90 curated samples >> 1,623 raw.",
  type="Insight",
  entities=["project/my-app"]
)