In Supermemory
You need best-in-class semantic recall — Supermemory tops LongMemEval, LoCoMo, and ConvoMem.
The memory + context API for the AI era — recall, RAG, and user profiles in one graph.
Comparison
nlqdb vs Supermemory
Pick Supermemory if you want a best-in-class memory API — fact extraction, hybrid recall, and connectors that top the memory benchmarks. Pick nlqdb if your agent also needs to run analytical queries (counts, group-bys, reports) over the structured rows it remembers.
The same goal, two ways.
> tasks each user completed per week this quarter, most recent first
In your HTML
<nlq-data goal="tasks each user completed per week this quarter, most recent first"></nlq-data>The aggregate slice over remembered rows is nlqdb's lane; Supermemory's lane is fuzzy recall of what was said, not GROUP BY over what was done.
What's different
Four dimensions that decide it
| Dimension | nlqdb | Supermemory | Note |
|---|---|---|---|
| Structured rows + typed columns | Supermemory stores memories in a vector graph; nlqdb stores typed rows in Postgres. | ||
| Natural-language recall / queries | |||
| Aggregations + reporting (COUNT, GROUP BY, JOIN) | Supermemory ranks and returns memories; it has no SQL aggregation over them. | ||
| Hybrid / semantic recall over unstructured text |
Show 4 more rows
| Dimension | nlqdb | Supermemory | Note |
|---|---|---|---|
| Connectors (Drive / Gmail / Notion / GitHub) | |||
| MCP server (agent-callable) | |||
| Auto-migration via NL ('add a `tier` column') | |||
| Open source / self-hostable | Supermemory is MIT + a one-binary local mode; nlqdb is FSL 1.1 (source-available) with no GA self-host container yet. |
shipped · partial · not shipped
When to choose nlqdb
- Your agent remembers structured rows it later counts, groups, and reports over in SQL.
- You want one MCP server that provisions Postgres, stores rows, and answers aggregations.
- The schema must evolve as the agent learns — 'add a `tier` column'.
- You need typed columns and exact answers, not similarity-ranked text snippets.
When to choose Supermemory
- You need best-in-class semantic recall — Supermemory tops LongMemEval, LoCoMo, and ConvoMem.
- Your memory is unstructured — documents, chat history, user facts as free text.
- You want connectors for Google Drive, Gmail, Notion, and GitHub out of the box.
- An MIT-licensed, self-hostable memory binary is on your shortlist.
Questions buyers ask
- Can I use Supermemory for recall and nlqdb for analytics over the same agent memory?
- Yes — they're complementary. Supermemory answers 'what did the user tell me about their project'; nlqdb answers 'how many tasks did the user complete each week this quarter'. Both sit behind one MCP-aware agent; nlqdb's MCP server exposes `nlqdb_query`, which materialises Postgres on first reference, so the structured side self-provisions.
- Does nlqdb beat Supermemory on memory benchmarks like LongMemEval?
- No, and it doesn't try to. Supermemory is purpose-built for semantic recall and leads LongMemEval, LoCoMo, and ConvoMem. nlqdb's lane is exact, aggregate queries — counts, group-bys, joins, reports — over the structured rows an agent stores, which those benchmarks don't measure.
- Is nlqdb a vector database?
- No. nlqdb is Postgres-first (ClickHouse for analytical engines in Phase 2). For semantic recall over unstructured text, Supermemory, Mem0, or pgvector are the right shape; nlqdb's contract is relational SQL over typed rows.
- How does my agent provision an nlqdb database autonomously?
- The MCP server exposes `nlqdb_query` — your agent calls it with a goal in English, the server materialises Postgres plus schema on first reference, and returns the answer bound to the agent's tenant. There's no separate create-DB verb to call first.
- Can nlqdb and Supermemory serve the same agent at once?
- Yes. Wire Supermemory's MCP server for fuzzy recall and nlqdb's for structured queries; the agent picks the right tool per question. Tenant-scoped `sk_live_*` keys keep each agent's nlqdb data isolated.
Try nlqdb in 30 seconds
No sign-in. The anonymous database lasts 72 hours; adopt it with one click if you keep it.
Start with a goal →