trivia

trivia

Semantic memory for Claude Code. Memorize facts, recall them by meaning, and let connections form automatically.

Quick Start

# Install (from source for now)
cargo install --path apps/cli

# As a Claude Code plugin
claude plugin add chrisdickinson/trivia
claude plugin install trivia

Basic Usage

# Store a fact
trivia memorize "project architecture" "Three-layer: API, service, storage. Each layer in its own crate."

# Recall by meaning
trivia recall "how is the code organized"

# Tag for organization
trivia memorize "auth flow" "OAuth2 PKCE with JWT refresh tokens" --tag backend --tag auth

As an MCP Tool

When installed as a Claude Code plugin, trivia is available as an MCP server. Claude can memorize and recall facts during conversations.

Features

• Semantic search via embeddings (AllMiniLM-L6-V2, 384-dim)
• Auto-linking — similar memories are linked automatically
• Auto-merging — very similar memories merge on creation
• Manual links — related, supersedes, derived_from
• Composite scoring — similarity + recency + frequency + link boost + ratings
• Tagging — categorize and filter memories
• Rating feedback — mark memories as useful/not to improve ranking
• Export/Import — markdown files with YAML frontmatter
• Web UI — browse, edit, search, graph visualization
• MCP server — Claude Code integration via stdin/stdout JSON-RPC

CLI Reference

trivia memorize <mnemonic> <content> [--tag <tag>...]
trivia recall <query> [--limit N] [--tag <tag>...] [--json]
trivia link <source> <target> [--link-type related|supersedes|derived_from]
trivia links <mnemonic>
trivia merge <keep> <discard>
trivia rate <mnemonic> --useful|--not-useful
trivia export <directory> [--tag <tag>...]
trivia import <directory>
trivia list-tags [--json]
trivia automerge [--threshold 0.25] [--dry-run]
trivia www [--port 3000]
trivia mcp

MCP Tools

Tool	Description
memorize	Store a fact with mnemonic, content, and optional tags
recall	Search by semantic similarity
rate	Provide useful/not-useful feedback
link	Create typed connections between memories
merge	Consolidate duplicate memories
export	Save memories to markdown files (optional tag filter)
import	Load memories from markdown files
list-tags	List all tags with counts

Configuration

Create a trivia.toml in your project root:

# Auto-add these tags to every memorize call
[memorize]
tags = ["my-project", "backend"]

# Boost these tags in recall scoring (not a filter — all memories still searchable)
[recall]
tags = ["my-project"]

# Default tag filter for export
[export]
tags = ["my-project"]

# Optional: override database path (default: ~/.claude/trivia.db)
# database = "/path/to/trivia.db"

Config discovery walks up from CWD (or CLAUDE_PLUGIN_ROOT) to find the nearest trivia.toml. CLI flags are additive with config tags.

Web UI

Start with trivia www and open http://localhost:3000. Features:

• Memory list with search and tag filtering
• Memory detail view with inline editing (including mnemonic rename)
• Link management
• Interactive merge
• Force-directed graph visualization

Architecture

crates/core/     — MemoryStore, Embedder, config, export/import
apps/cli/        — CLI binary (`trivia`), web server, MCP server
apps/cli/www/    — React + TypeScript web UI (embedded at build time)

SQLite with sqlite-vec for vector search. Embeddings via fastembed (AllMiniLM-L6-V2).

Environment Variables

• TRIVIA_DB — database path (overrides config and default)
• CLAUDE_PLUGIN_ROOT — plugin root for config discovery

License

MIT