GetProfile

GetProfile solves a specific, real problem well: how to give an LLM agent durable, structured knowledge of a user without stuffing a growing message history into every prompt. The proxy sits between your agent and the LLM provider, enriches each request with a system message built from the user's profile summary, typed traits, and relevant memories, then updates that profile in the background after the response. That design keeps prompts lean and predictable — the project is explicit that it avoids overloaded context windows and black-box memory behavior. The strongest part is the customizable trait schema. You decide what traits matter for your domain, how they are extracted from interactions, and how they are injected into prompts, with example trait definitions in the docs. Traits carry confidence scores and memories carry importance levels, so the profile is structured rather than a dumped text blob. It is Apache 2.0 licensed, built on Hono, and stores everything in your own PostgreSQL database, which is the right fit for privacy-conscious teams that need auditability. The honest weaknesses: it is self-hosted, so you need Docker and PostgreSQL fluency, and there is no zero-configuration option yet — GetProfile Cloud is announced but not available today. It is text-only, so multi-modal memory is out of scope. And operating it means managing infrastructure, even if the proxy itself is lightweight and scales with Docker Compose. If you want a managed memory layer, this is not it yet; if you want ownership and transparency, it is a credible choice.