CortexaDB

Positioned as "SQLite for AI Agents," CortexaDB addresses the persistent memory problem for AI Agents after context window overflow. The core engine is written in Rust (cortexadb-core) and exposes high-performance Python bindings (cortexadb-py) via maturin, managed as a Cargo workspace.

Retrieval Engine#

• Vector Search: Native semantic similarity retrieval using HNSW algorithm (backed by USearch library) for sub-millisecond approximate nearest neighbor search.
• Graph Relationships: Supports creating and querying relationship edges between memory entries (e.g., relates_to), mixable with vector search in the same query.
• Temporal Indexing: Built-in time-dimension indexing supporting time-decay/recency-based retrieval.
• Hybrid Retrieval: A single query can leverage all three signals—semantic similarity, structural relationships, and temporal decay.

Storage & Performance#

• WAL (Write-Ahead Log) backed storage ensures zero data loss.
• Batch writes claimed at 5,000+ chunks/second (README claim, no independent benchmark data found).

Usability#

• Zero-dependency / Single-file: No server process needed, SQLite-like embedded experience.
• Fluent Query Builder: Chain-style Python API (.query().limit().use_graph().execute()).
• Embedder Abstraction: Inject external embedding models via provider interface (only OpenAIEmbedder shown in docs; other model support unconfirmed).
• Document Parsing (optional): Install cortexadb[docs,pdf] for PDF/DOCX parsing and import.

Privacy & Deployment#

• Fully local execution; data never leaves the user's machine.
• Cross-platform: Windows, Linux, macOS.
• Current stable version v1.0.1, dual-licensed under MIT + Apache-2.0.

Quick Start#

from cortexadb import CortexaDB
from cortexadb.providers.openai import OpenAIEmbedder

db = CortexaDB.open("agent.mem", embedder=OpenAIEmbedder())
mid1 = db.add("The user prefers dark mode.")
mid2 = db.add("User works at Stripe.")
db.connect(mid1, mid2, "relates_to")
hits = db.query("What are the user's preferences?").limit(5).use_graph().execute()

Unconfirmed Items#

• Batch write performance lacks independent benchmark verification.
• Documentation site cortexa-db.vercel.app has SSL certificate issues; full docs cannot be verified.
• PyPI page protected by JS challenge; package details unverifiable.
• use_graph() graph traversal algorithm (BFS/DFS/PageRank etc.) and depth limits undocumented.
• Whether cortexadb-core is published to crates.io is unclear.