An open-source 'super memory' for AI agents was posted — multi-agent RAG + hybrid wiki search + transparency score. The missing piece in most memory systems: a live search API that keeps the wiki fresh. Five ranked for memory grounding.
Top Pick
Scavio's multi-platform search provides the freshest grounding data across Google, Reddit, and YouTube for agent memory wikis.
Full Ranking
#1Our Pick
Scavio
Multi-source memory grounding (Google + Reddit + YouTube)
Pros
- Multiple platforms catch different freshness signals
- Reddit threads provide human-written context
- Credit-based scales with memory refresh frequency
Cons
- Returns raw SERP, not pre-processed
#2
Tavily
Simple web grounding for memory refresh
Pros
- Clean JSON
- Research API for deeper grounding
Cons
- Web only
- No Reddit/YouTube
#3
Exa
Semantic memory search
Pros
- Neural search finds conceptually related content
- Contents included
Cons
- No structured platform data
- Different from keyword grounding
#4
Brave Search API
Budget memory refresh
Pros
- Cheapest per-query
- Independent index
Cons
- Web only
- No platform-specific data
#5
Perplexity Sonar
Pre-synthesized memory updates
Pros
- Returns synthesized answers
- Saves post-processing
Cons
- Credit wipe reports (May 2026)
- Token-based pricing less predictable
Side-by-Side Comparison
| Criteria | Scavio | Runner-up | 3rd Place |
|---|---|---|---|
| Data sources | Google + Reddit + YouTube + Amazon + Walmart | Web (Tavily) | Web (Exa, semantic) |
| Memory refresh cost (100 entries) | $0.50 | $0.50 (Tavily) | $0.70 (Exa) |
| Pre-synthesized answers | No (pair with LLM) | Research API (limited) | No |
| Reddit context | Yes (Reddit endpoint) | No | No |
Why Scavio Wins
- Agent memory needs multi-source grounding. A fact verified on Google AND corroborated on Reddit has higher confidence than a single-source check.
- Sonar's pre-synthesized answers save post-processing but the credit wipe report undermines production reliability. For persistent memory systems, predictable billing matters more.
- Exa's semantic search is genuinely better for finding conceptually related content to memory entries. For memory deduplication and similarity search, Exa wins.
- The memory refresh pattern: cron job → stale entries → search API → update wiki → transparency score. All five APIs work for this; the differentiator is source diversity.
- Reddit as a freshness signal: human-written threads about a topic often surface sooner than formal web articles. Scavio's Reddit endpoint provides this signal; others do not.