Back to Projects

pg-agent-memory

Stateful AI agent memory layer for PostgreSQL with pgvector. TypeScript-first with intelligent context management and zero-cost embeddings.

Local

Embeddings

Multi-model

Token counting

PostgreSQL

pgvector native

Core Features

Local Embeddings

Zero-cost semantic search using local Sentence Transformers

PostgreSQL Native

Uses existing PostgreSQL infrastructure with pgvector

Architecture

Memory Layer

TypeScript-first API with intelligent context management and compression

Multi-Model Support

Universal tokenizer for OpenAI, Anthropic, DeepSeek, Google, Meta

TypeScriptPostgreSQLpgvectorSentence TransformersVitest

Design Decisions

PostgreSQL-First

Leveraged existing infrastructure with pgvector instead of separate vector databases

Local Embeddings

Zero-cost embedding generation eliminating API dependencies

Universal Tokenization

Provider-specific token counting for accurate cost estimation