pumaDB

pumaDB is an AI developer tool that gives AI agents a shared, persistent memory layer without requiring any database, vector store, or RAG stack to set up. It solves the common problem of agents losing useful context between sessions, tools, and chats by offering one durable place to save and reuse notes, facts, preferences, project context, transcripts, and task state. Agents connect through two access methods: a hosted MCP (Model Context Protocol) server at api.pumadb.ai/mcp for agent clients like ChatGPT, Claude, and Codex, or a server-side REST API for application code. pumaDB stores multiple memory types — skills markdown, project conventions, user preferences, research clippings, task state, and typed safe memory — and adds safety controls including scoped table limits, rate limiting, and natural-language edit filtering. It keeps automatic version history with 30-day retention, and every update or delete archives the prior row so content can be restored. Documented account limits include 20 tables, 1,000 rows per table, and 25 MB of storage, with rate limits of 30 writes and 60 reads per minute per key, and OAuth authentication for the hosted MCP endpoint. As an AI developer tool for agent memory, it targets developers who want a persistent memory backend without managing infrastructure.

Giving developers a persistent, shared memory backend for AI agents without setting up a database, vector store, or RAG stack