An open-source 'super memory' for AI agents was posted — multi-agent RAG + hybrid wiki search + transparency score. This tutorial builds the search-grounded memory layer.
Prerequisites
- Scavio API key
- Git repository for the wiki
- LLM API key
Walkthrough
Step 1: Design the wiki structure
Markdown files organized by topic with metadata frontmatter.
# wiki/topics/search-api-pricing.md
# ---
# topic: Search API Pricing
# last_verified: 2026-05-03
# confidence: high
# sources: [scavio.dev, serpapi.com, tavily.com]
# ---
# Scavio: $0.005/query, $30/mo for 7K credits
# SerpAPI: $75/mo for 5K searches
# ...Step 2: Build the refresh function
Search API verifies and updates stale entries.
Python
import requests, os, datetime
H = {'x-api-key': os.environ['SCAVIO_API_KEY']}
def refresh_entry(topic, current_content):
results = requests.post('https://api.scavio.dev/api/v1/search', headers=H,
json={'platform': 'google', 'query': f'{topic} 2026 current'}).json()
# LLM compares current wiki content vs fresh search results
# Updates only if facts changed
# Logs the diff for transparencyStep 3: Add transparency scoring
Each entry gets a confidence score based on recency and source count.
Python
def transparency_score(entry):
days_since_verified = (datetime.date.today() - entry['last_verified']).days
if days_since_verified < 7: return 'high'
if days_since_verified < 30: return 'medium'
return 'stale'Step 4: Schedule refresh cron
Daily cron refreshes stale entries.
Bash
# crontab: 0 3 * * * python refresh_wiki.py
# Refreshes entries with confidence < 'high'
# Budget: ~50 queries/day = $0.25/dayStep 5: Git-track changes
Every refresh is a git commit for full diff history.
Bash
# After refresh:
git add wiki/
git commit -m "refresh: $(date +%Y-%m-%d) - N entries updated"
# Full diff history in git logPython Example
Python
# Agent memory wiki: markdown files + search grounding + git history
# The wiki IS the memory. Git IS the audit trail.
# Search API keeps it fresh. LLM decides what changed.JavaScript Example
JavaScript
// Same pattern in JS/TS with file system operations.Expected Output
JSON
Git-tracked markdown wiki with daily search-grounded freshness verification, transparency scores, and full diff history.