You want a memory API or second brain.
supermemory is the natural fit when you need a general-purpose memory layer, a second brain for personal knowledge, or an API you can integrate into your own agent or application.
Source-linked comparison
supermemory is a general memory layer and second brain. ContextStream is project memory for AI coding agents: search, decisions, lessons, plans, handoffs, and MCP access for Cursor and Claude Code.
10,000 credits every month. Full platform access. No credit card.
At a glance
Sources in the table below| Dimension | ContextStream | supermemory |
|---|---|---|
| Product center | Project memory for AI coding agents | A general memory layer and second brain |
| MCP access | Yes | Yes |
| What is remembered | Decisions, lessons, plans, handoffs and code | Source documents and extracted memories |
| LongMemEval-S | 96.8%ContextStream-published | 85.4%vendor-published |
Product-authored comparison · supermemory details link to their own documentation · vendor-published scores cited with source
01Capability comparison
This table links to supermemory’s own documentation for each detail. Follow the source links for their latest features.
| Dimension | ContextStream | supermemory |
|---|---|---|
| Product center | A complete AI workspace with agent orchestration, coordination, decision learning, precise code search and full code graphs, also available through MCP. | A memory platform for AI tools and applications, with shared spaces, MCP access and interactive memory workflows. supermemory docs ↗ |
| MCP access and editors | YesMCP server connects Cursor, Claude Code, Cline, and Continue to project memory. Code search and structured tools included. | YesHosted MCP with OAuth connects compatible assistants, including Claude Code, Cursor and Codex. MCP documentation ↗ |
| What is remembered | Decisions, lessons, plans, handoffs, code search results, dependency graphs, and structured project knowledge. | Source documents and extracted memories, with search, save, forget and file upload workflows. Memory model ↗ |
| Code search and graph | Code search plus full dependency, call-path, impact and quality analysis. Published best-answer Recall@10 is 98.77% (useful-file: 100%) for ContextStream auto in production on October 27, 2026. | The linked MCP documentation describes semantic memory search and a graph of documents and memories. Evaluate code-specific analysis against your project requirements. Search docs ↗ |
| Team and project scope | Workspaces, projects, team roles, and project-level access controls. Memory scoped to the right project and team members. | Users can organize memories and control access through spaces. Spaces docs ↗ |
| LongMemEval-S accuracy | 96.8%96.8% on LongMemEval-S (ContextStream-published, official GPT-4o judge; 96.0% on a held-out half). View benchmarks → | 85.4%85.4% on LongMemEval-S (vendor-published by supermemory). supermemory benchmarks ↗ |
02Decision guide
supermemory is the natural fit when you need a general-purpose memory layer, a second brain for personal knowledge, or an API you can integrate into your own agent or application.
ContextStream is built for coding projects. Code search, decisions, lessons, plans, handoffs, dependency graphs, and continuity across Cursor and Claude Code sessions should arrive as one layer.
CursorThe policy is one refresh retry, then sign out. It was decided on Feb 10 because a second retry caused sign-in loops, and it lives in refresh-coordinator.ts, lines 112–148. I'll keep that behaviour and read those lines before editing.
One layer for a coding project. The decision, the lesson and the code behind it come back in the same Cursor session.
Pricing
Starter includes the full platform and 10,000 credits every month. Pro is $19/month with 25,000 credits. Elite is $49/month with 100,000 credits. Every plan includes the same agent orchestration, memory, search and code intelligence.
Published evidence
ContextStream reports 96.8% accuracy on LongMemEval-S and 98.77% best-answer Recall@10 in its published code-search evaluation. These are ContextStream measurements on ContextStream tasks.
Connect Cursor or Claude Code, index one repository, and test whether the next session retrieves a real decision and the code it affects.