GoodMem

GoodMem is a memory layer for AI agents that handles ingestion, permissioned retrieval and automatic tuning on top of PostgreSQL with pgvector. It can run in its cloud, a VPC, on-prem or air-gapped, and is aimed at teams building production agents.

GoodMem is memory infrastructure for AI agents. It connects agents to a company's documents, data and their own interaction history, then serves the right context on demand instead of stuffing prompts. The pitch centers on lower token use, retrieval that is measured rather than assumed, and the option to keep everything inside the customer's own network. In practice, documents are ingested and stored in a PostgreSQL database with pgvector, so there is no proprietary store to migrate away from. Every page of a PDF, Word, PowerPoint or Excel file is also rendered to an image at ingestion, which lets a multimodal model read tables, charts and drawings that plain text extraction flattens. Access is enforced through scoped API keys: a request for a space the key cannot read is refused with a 403 before any search runs, and requests can be logged per the customer's policy. The Retrieval Optimizer, available in GoodMem Cloud, benchmarks combinations of embedders and rerankers (open-weight or API) on the customer's own queries and reports confidence ranges, so teams can trade a little quality for cost or latency. An Auto-Optimizer fine-tunes retrieval models over time. The audience is engineering teams building production agents, particularly in regulated industries or places where data cannot leave the premises. GoodMem can be self-hosted for free, run as a managed cloud service, or deployed on-prem, in a VPC or fully air-gapped under an enterprise plan. It works with hosted providers such as OpenAI and AWS Bedrock as well as local stacks like Ollama, vLLM, TEI and OpenRouter, and it is available through ChatGPT, Claude and Google ADK. The company also offers custom model training through its PAIR program for teams that want a smaller self-run model. Compared with a general vector database or a hand-built RAG pipeline, GoodMem bundles ingestion, permissions, evaluation and tuning into one layer. That breadth suits teams that want governance and evidence, but it is more than a hobby project needs.