We spent weeks digging through GitHub issues, benchmark papers, developer forums, and actual codebases to build this guide. No affiliate links, no sponsored placements — just what each tool actually does, who should use it, and what nobody tells you until you're three days into integration.
Last updated February 2026 · 5 tools covered
How we think about AI memory
We use a three-tier framework: session memory (dies when you close the tab), persistent memory (remembers per user, per tool), and shared memory (team-wide, always current, works across every tool).
Most tools in this space are building Tier 2. We think Tier 3 is where the real value is for engineering teams. Read the full framework →
| Tool | Best for | MCP | Free tier | Starts at |
|---|---|---|---|---|
| Mem0 | memory SaaS running in production this week | Yes | 10 | $19/mo (Starter) |
| Zep | "when did we learn this" matters as much as "what | Yes | API access with lowest rate limits | Credit-based: 1 Episode = 1 credit |
| Cognee | multi-hop reasoning, connecting facts across multi | Yes | Fully open source (Apache 2.0 | C |
| Letta | from scratch where memory is a first-class concern | Yes | 5 | $20/mo (Pro): 20,000 credits, unlimited agents |
| LangMem | deep in LangChain/LangGraph who want memory withou | No | Open source (MIT). Managed service with free long-term memory launching. | LangSmith (observability) has separate pricing tiers |
The fastest path from zero to deployed memory
Mem0 is the safe pick if you need your chatbot to remember users. The API is clean, latency is the fastest we've seen (p50: 148ms), and it handles the messy stuff (deduplication, conflict resolution, decay) so you don't have to.
Temporal knowledge graph for agents that need to know when
Zep is the most interesting tool in this space, at least technically. The temporal knowledge graph is a novel idea.
Knowledge graphs for AI that needs to connect the dots
Cognee is what you reach for when you actually need AI to understand relationships between things, not just retrieve similar text. Their HotPotQA benchmark score of 0.
Agents that manage their own memory (formerly MemGPT)
Letta is the most ambitious project in this space. The idea that agents should manage their own memory (deciding what to remember, forget, and retrieve) is compelling on paper.
Memory SDK for teams already living in LangChain
LangMem makes sense if and only if you're already committed to LangGraph. The three memory types (semantic, episodic, procedural) are well-designed, and the Prompt Optimizer, which automatically improves agent behavior based on conversation patterns, is something nobody else offers.
Notice something missing?
Every tool on this page is Tier 2 — persistent memory scoped to individual users or agents. None of them solve the team problem: shared, always-current memory that works across Cursor, Claude, and your custom agents simultaneously.
We're building that. It's called Knowledge Plane. Join the beta →