LLM Memory Ecosystem: Products & Open Source Projects
Last Updated: 2025-12-18
Research Methodology: Combined Claude Code analysis + Gemini CLI internet search (Dec 2025)
0. Market Analysis Summary
Research Status in This Project
| Project |
Status |
Research File |
| Mem0 |
✅ Researched |
mem0.research.md |
| Letta |
✅ Researched |
letta.research.md |
| Cursor |
✅ Researched |
cursor.research.md |
| Augment Code |
✅ Researched |
augmentcode.research.md |
| Zep + Graphiti |
✅ Researched |
graphiti.research.md |
| Continue |
✅ Researched |
continue.research.md |
| Qdrant |
✅ Researched |
qdrant.research.md |
| Chroma |
✅ Researched |
chroma.research.md |
Research Priority Recommendations
| Priority |
Project |
GitHub Stars |
Funding |
Reason |
| 🔥 1 |
Zep + Graphiti |
~21.2k |
$2.3M (YC) |
Temporal knowledge graph - new direction in 2025 |
| 🔥 2 |
Continue |
30.4k |
$65M Series A |
Open source coding assistant market leader |
| ⭐ 3 |
Qdrant |
27.7k |
$37.8M |
Production-scale vector DB leader |
| ⭐ 4 |
Chroma |
24k |
$36M |
Developer experience leader, Rust rewrite v1.0 (Mar 2025) |
| ○ 5 |
Tabby |
11k |
$7.2M |
Self-hosted alternative (Continue is leader) |
| ○ 6 |
Void |
- |
- |
Privacy-first, new project |
Key 2025 Insight
Industry shift: From pure vector similarity → Knowledge Graphs + Temporal Awareness
Graphiti's 21.2k stars (vs Zep core's 3.8k) shows massive developer interest in graph-based memory.
1. Memory Frameworks
| Project |
Type |
GitHub Stars |
Funding |
Core Technology |
2025 Updates |
Link |
| Mem0 |
Open Source |
~23k |
$2.2M |
LLM-driven fact extraction + Vector/Graph dual storage |
Mem0g - Graph database for relational reasoning |
GitHub |
| Letta (MemGPT) |
Open Source |
~15k |
$10M |
Three-tier memory + Agent self-editing prompts |
Supports "infinite context" chatbots |
GitHub |
| Zep |
Open Source |
~3.8k (core) |
$2.3M (YC) |
Long-term memory + Auto-summarization + Entity extraction |
Graphiti temporal knowledge graph engine |
GitHub |
| Graphiti |
Open Source |
~21.2k |
(see Zep) |
Temporal knowledge graph engine |
Breakout hit - massive adoption in 2025 |
GitHub |
| LangChain Memory |
Open Source |
(part of LangChain) |
$35M+ |
Multiple Memory types (Buffer, Summary, VectorStore) |
Most complete ecosystem, but generic |
GitHub |
| LlamaIndex |
Open Source |
~38k |
$19M |
Data indexing framework, multiple storage backends |
Best for RAG scenarios |
GitHub |
| Motorhead |
Open Source |
~0.5k |
- |
Memory server with Redis backend |
Lightweight option, less active |
GitHub |
Selection Guide
| Use Case |
Recommendation |
| Infinite context chat |
Letta |
| Smart user profiles |
Mem0 or Zep |
| Temporal-aware memory |
Zep (Graphiti) |
| Quick prototyping |
LangChain Memory |
2. AI Coding Assistants (Codebase Indexing)
| Project |
Type |
GitHub Stars |
Funding |
Core Technology |
Market Position |
Link |
| Cursor |
Commercial |
- |
$400M+ (Series B) |
AST + Symbol graph + Semantic embeddings |
Commercial benchmark, custom embedding models |
cursor.com |
| Augment Code |
Commercial |
- |
$252M |
Context Engine + Real-time personal index |
Enterprise leader, per-developer index |
augmentcode.com |
| Continue |
Open Source |
30.4k |
$65M Series A |
Customizable Context Providers |
Open source leader, $500M valuation |
GitHub |
| Tabby |
Open Source |
11k |
$7.2M |
RAG + repo-context indexing |
Self-hosted Copilot alternative |
GitHub |
| Aider |
Open Source |
~25k |
- |
Git-aware multi-file editing |
CLI tool leader |
GitHub |
| Sourcegraph Cody |
Open Source |
~3k |
$225M (company) |
Enterprise code search |
Cross-repository search |
GitHub |
| Void |
Open Source |
~5k |
- |
Cursor-like features |
Privacy-first, new entrant |
voideditor.com |
| CocoIndex |
Open Source |
~1k |
- |
Tree-sitter syntax-aware chunking |
Niche: syntax-aware indexing |
GitHub |
Selection Guide
| Use Case |
Recommendation |
Notes |
| Open source IDE extension |
Continue |
Market leader, $65M funding |
| Privacy-first |
Void |
Runs locally with your own API keys |
| Self-hosted enterprise |
Tabby |
$7.2M funding, enterprise focus |
| CLI workflow |
Aider |
Git-aware, terminal-native |
| Syntax-aware indexing |
CocoIndex |
Tree-sitter chunking by function/class |
3. Vector Databases
| Project |
Language |
GitHub Stars |
Funding |
2025 Features |
Market Position |
Link |
| Qdrant |
Rust |
27.7k |
$37.8M |
High performance + Payload filtering |
Production leader, 2025 AI TechAward, Cloud Inference (Jul 2025) |
GitHub |
| Chroma |
Rust (v1.0) |
24k |
$36M |
Zero config, in-memory/persistent |
DevEx leader, Rust rewrite v1.0 (Mar 2025), 90k+ integrations |
GitHub |
| Weaviate |
Go |
~13k |
$50M |
Hybrid search (BM25 + vector) |
Best for code search (exact variable matching) |
GitHub |
| Milvus |
Go |
~32k |
$113M |
Distributed, billion-scale vectors |
Large-scale production leader |
GitHub |
| pgvector |
C |
~13k |
- |
PostgreSQL extension, ACID |
Unified database for Postgres users |
GitHub |
| Pinecone |
Commercial |
- |
$138M |
Fully managed, zero ops |
Managed service leader |
pinecone.io |
Selection Guide
| Use Case |
Recommendation |
Notes |
| Production performance |
Qdrant |
$37.8M funding, Rust-based, lowest latency |
| Quick prototyping |
Chroma |
DevEx focus, zero config, Rust v1.0 |
| Code search |
Weaviate |
Hybrid search for exact variable names |
| Unified database |
pgvector |
No new infrastructure if using Postgres |
| Billion-scale |
Milvus |
$113M funding, distributed architecture |
4. Graph Databases (Knowledge Graphs)
| Project |
Type |
GitHub Stars |
Highlights |
LLM Memory Use Case |
Link |
| Graphiti |
Open Source |
~21.2k |
Zep's temporal knowledge graph engine |
LLM-native, designed for agent memory |
GitHub |
| Neo4j |
Commercial/Community |
~14k |
Most mature graph database |
General-purpose, not LLM-specialized |
neo4j.com |
| Memgraph |
Open Source |
~1k |
In-memory first, high performance |
Real-time graph analytics |
GitHub |
| Kuzu |
Open Source |
~1.5k |
Embedded graph database |
Local/embedded use cases |
GitHub |
Note: Graphiti's 21.2k stars significantly outpace traditional graph DBs for LLM use cases, indicating strong developer preference for LLM-native solutions.
5. Embedding Models
| Project |
Type |
Highlights |
Link |
| OpenAI Embeddings |
Commercial |
text-embedding-3-small/large |
openai.com |
| Voyage AI |
Commercial |
Code-specialized embeddings |
voyageai.com |
| BGE |
Open Source |
BAAI, multilingual support |
HuggingFace |
| Nomic Embed |
Open Source |
Long context support |
GitHub |
| Jina Embeddings |
Open Source |
8K context window |
jina.ai |
6. Key Trends in 2025
| Trend |
Description |
Representative Project |
| Graph-Enhanced Memory |
From pure vector similarity to relational reasoning |
Mem0g |
| Temporal Knowledge |
Time-aware context management |
Zep Graphiti |
| Privacy-First |
Open source alternatives to proprietary tools |
Void |
| Syntax-Aware Chunking |
Tree-sitter chunking by function/class |
CocoIndex |
| Hybrid Search |
Keyword + vector search combined |
Weaviate |
| Custom Embeddings |
Training from agent session traces |
Cursor |
| Real-Time Personal Index |
Per-developer independent index |
Augment Code |
7. Recommended for Further Research
Based on existing research (mem0, letta, augment, cursor) and market analysis:
High Priority (New Technical Direction or Market Leader)
| Project |
GitHub Stars |
Funding |
Why Research |
| Zep + Graphiti |
21.2k |
$2.3M (YC) |
Temporal knowledge graph - 2025's new direction for agent memory |
| Continue |
30.4k |
$65M |
Open source IDE market leader, $500M valuation |
Medium Priority (Infrastructure Leaders)
| Project |
GitHub Stars |
Funding |
Why Research |
| Qdrant |
27.7k |
$37.8M |
Production vector DB leader, Rust performance |
| Chroma |
24k |
$36M |
DevEx leader, Rust v1.0 rewrite |
Optional (Niche or Less Differentiated)
| Project |
GitHub Stars |
Notes |
| Tabby |
11k |
Self-hosted alternative, but Continue is leader |
| Void |
~5k |
Privacy-first, too new to assess |
| CocoIndex |
~1k |
Tree-sitter niche, small community |
8. Tech Stack Examples
Scenario A: Personal Project / Prototype
Embedding: OpenAI text-embedding-3-small
Vector DB: Chroma (local)
Memory: LangChain ConversationBufferMemory
Scenario B: Production AI Assistant
Embedding: Voyage AI (code-specialized)
Vector DB: Qdrant (high performance)
Memory: Mem0 (user profiles) + Letta (long conversations)
Graph: Neo4j (relational reasoning)
Scenario C: Self-Hosted Code Assistant
Codebase Index: Tabby / Continue
Embedding: BGE / Nomic (local)
Vector DB: pgvector (unified database)
LLM: Ollama (local)
References