Architecture

How Multiplist's semantic memory works — the concepts behind the 19 public tools.

Multiplist organizes knowledge into a layered structure built for AI-native retrieval with full provenance.

# Sources

The atomic unit of imported content. Conversations, documents, transcripts, notes. Every source carries:

• Raw content (the original text)
• Metadata (title, source type, tags, external IDs for dedup)
• An abstract (generated automatically unless disabled)
• A pipeline status (raw → scanning → extracting → complete)

# Seeds

Structured insights extracted from sources. Nine categories:

• Decisions — choices made, directions locked
• Frameworks — mental models, systems, structured approaches
• Golden Passages — phrases worth keeping exactly as spoken
• Definitions — terms coined, concepts named
• Exemplars — quality examples worth replicating
• Actions — tasks, next steps, follow-ups
• Questions — open loops, unresolved threads
• Offers — proposed next steps, invitations
• Emergence — patterns forming across topics

Every seed records its source reference and exact character positions — that's what makes Multiplist an IP Vault with Citations.

# Containers & Studios

• Containers — top-level domains for organizing studios. Client work, projects, long-running initiatives.
• Studios — focused workspaces within containers. Each studio has pinned sources and a card deck for active work.

# Extraction

The semantic engine that reads sources and produces categorized, cited seeds. Runs automatically on create_source unless you opt out; can be re-run with trigger_extraction against a specific skill or category set.

# Skills

Custom extraction recipes. Skills tell Multiplist what to look for in your content beyond the default nine categories. Each skill defines a schema, trigger conditions, and extraction priorities. Skills are matched against sources at extraction time and produce seeds tagged with the skill that generated them.

# Research briefs

Long-running synthesis across vault sources. Dispatched via request_brief, polled via get_brief_status. Completed briefs are saved as their own sources — they become searchable alongside everything else.

# Annotations & marginalia

Two ways to write back to a source:

• Annotations — the author's own voice. Real-time notes, decisions, insights.
• Marginalia — later-mind marks. Corrections, tensions, evolutions, commentary. Records the scribe identity and mark type so cross-AI provenance stays clean.