If you've been searching for "AI memory for agents" or "knowledge extraction API," you've probably encountered both Supermemory and KnowledgeSDK. They appear in similar contexts — developer infrastructure for AI — and they overlap on a few surface features like hybrid search and MCP server support. But they solve fundamentally different problems, and conflating them leads to choosing the wrong tool.
This is an honest comparison. We built KnowledgeSDK, so take that into account — but we've tried to represent Supermemory accurately based on their public documentation and announcements.
Background
Supermemory raised a $3M Series A and has positioned itself as "the memory layer for AI agents." Their core product is a persistent memory store that combines vector and graph representations to store what an agent has learned across sessions. They're ranked #1 on MemoryBench, the primary benchmark for agent memory systems. Enterprise customers include BrowserStack, Nissan, and Zscaler. Their pitch: your AI agent should remember user preferences, past conversations, and accumulated knowledge the same way a human assistant does — persistently, across sessions.
KnowledgeSDK is a web-first knowledge extraction and search API. The core loop: give it a URL, it extracts clean, structured content from that page (including JavaScript-rendered content), indexes it in your private collection, and makes it instantly searchable via hybrid semantic + keyword search. The pitch: your AI agent should know what's on any website, right now, without you scraping it yourself.
The Core Difference
The clearest way to state it:
Supermemory = what your agent remembers from past interactions
KnowledgeSDK = what's published on the web right now
Supermemory is episodic and semantic memory: "This user prefers Python over TypeScript." "The last time we discussed this customer's contract, they mentioned a Q3 deadline." "This user's coding style uses 2-space indentation."
KnowledgeSDK is web intelligence: "Here's everything on this competitor's pricing page." "Here's the latest documentation from this library." "Here are the 50 pages on this company's website, indexed and searchable."
One stores what happened in your system. The other surfaces what exists on the internet.
When Supermemory Wins
If your primary use case involves any of the following, Supermemory is probably the right tool:
Persistent user memory: Your AI assistant needs to remember that a user prefers formal tone, works in the healthcare industry, and mentioned their main concern is HIPAA compliance. This is agent memory — accumulated across sessions, specific to one user.
Conversation history with semantic search: You need to retrieve relevant past conversations. "Did we discuss this feature request before?" requires searching your own system's history, not the web.
Agent state across sessions: A coding agent that remembers what files it's modified, what decisions were made, and what the user said they were working toward. Supermemory's vector-graph combination handles this well.
Sub-300ms memory recall: Supermemory has invested heavily in retrieval latency. If your agent needs to recall memory in real time during a user interaction, their infrastructure is optimized for this.
When KnowledgeSDK Wins
If your primary use case involves any of the following, KnowledgeSDK is the right tool:
Extracting live web content: Your agent needs to read a competitor's website, a job posting, a news article, or any publicly available URL. KnowledgeSDK handles JavaScript rendering, anti-scraping measures, and content cleaning.
Building a private knowledge base from URLs: You want your agent to "know" a specific set of websites — your company's documentation, partner sites, industry publications — without building a scraping and indexing pipeline yourself.
Detecting when web content changes: KnowledgeSDK supports webhooks for change detection. When a monitored page updates, you receive a notification and the new content is automatically re-indexed. Supermemory has no equivalent for external web content monitoring.
Searching web content at query time: Your agent needs to answer questions grounded in specific websites. POST /v1/search returns ranked chunks from your extracted content in milliseconds.
Feature Comparison
| Feature | Supermemory | KnowledgeSDK |
|---|---|---|
| Memory storage (agent interactions) | Yes — core product | No — not the use case |
| Web content extraction | Partial (connector-based) | Yes — core product |
| Hybrid search (semantic + keyword) | Yes | Yes |
| MCP server | Yes | Yes |
| Webhooks / change detection | No | Yes |
| Node.js SDK | Yes | Yes (@knowledgesdk/node) |
| Python SDK | Yes | Yes (knowledgesdk) |
| MemoryBench rank | #1 | N/A (different category) |
| Pricing | Free / $19/mo (Hobby) | Usage-based |
| Self-hosted option | Open-source core available | Managed API |
| Multimodal memory | Yes | Partial (screenshots via /v1/screenshot) |
| Enterprise customers | BrowserStack, Nissan, Zscaler | — |
The Hybrid Architecture: Using Both
The most capable AI agent architectures use both types of infrastructure simultaneously — a memory layer and a knowledge layer. They're not competing for the same slot.
Consider a customer success AI agent:
Memory layer (Supermemory): Remembers every customer interaction, their product tier, past issues, stated preferences, renewal dates, and escalation history. When a customer opens a chat, the agent recalls who they are and what matters to them.
Knowledge layer (KnowledgeSDK): Knows your product documentation, competitor positioning, pricing pages, and industry news — all indexed and searchable. When a customer asks "how does your product compare to [Competitor]?" the agent searches the knowledge layer and answers with current information.
Neither tool can replace the other in this architecture. Supermemory can't tell you what's on a competitor's website today. KnowledgeSDK doesn't store what a specific user said in their last three support sessions.
// Simplified agent retrieval combining both layers
async function agentContext(userId: string, query: string) {
const [userMemory, webKnowledge] = await Promise.all([
// Supermemory: what do we know about this user?
supermemory.search({ userId, query, limit: 3 }),
// KnowledgeSDK: what does the web say about this topic?
fetch("https://api.knowledgesdk.com/v1/search", {
method: "POST",
headers: { "x-api-key": process.env.KNOWLEDGESDK_API_KEY! },
body: JSON.stringify({ query, limit: 5 }),
}).then((r) => r.json()),
]);
return {
userContext: userMemory.results,
webContext: webKnowledge.results,
};
}
Honest Verdict
Supermemory is the right choice when your agent's core value comes from remembering and learning from interactions over time. If your agent gets smarter about a user the more it interacts with them, you need Supermemory. Their MemoryBench #1 ranking is meaningful — they've invested deeply in the hard problems of agent memory: deduplication, memory consolidation, fast retrieval.
KnowledgeSDK is the right choice when your agent's core value comes from knowing what's currently on the web. If your agent answers questions about specific websites, monitors content for changes, or needs to reason about publicly available information, that's the knowledge layer — and that's what KnowledgeSDK is built for.
The tools are complementary. Your agent can have both a memory layer and a knowledge layer simultaneously. Choose based on your primary use case, and consider whether your application actually needs both.
If you're still uncertain: agents that serve individual users over time tend to need Supermemory. Agents that analyze external information (competitive intelligence, market research, documentation Q&A) tend to need KnowledgeSDK. Many production agents need both.