Jumbo

Jumbo is a command line tool that gives AI coding agents persistent project memory and structured context, so each session starts with the decisions, constraints, invariants and architecture it needs instead of being retaught the project. It hooks into the agent session automatically, collects insights as they arise during the workflow, and serves the relevant context back to the agent without anyone hand maintaining Markdown files. It is designed for developers who work with agents such as Claude Code and want better first drafts, longer sessions without context rot, and freedom to switch tools. The workflow has three steps: the developer defines goals from the command line or through a guided terminal wizard, Jumbo binds past memories to the goal and generates a tailored spec, and the agent is pointed at the goal. When an agent goes off track, the correction is captured once and applied to related future goals. Several agents can run in parallel, matched to capability and cost, while sharing the same context. Jumbo works with any harness and any model, is built on AGENTS.md and the open agent skill standard, and keeps all data in a local folder inside the project with no network calls, no API keys and no login. Jumbo is free and open source for single developer use, installed as a global npm package. No paid plan or team pricing is published. The source code is hosted on GitHub under the jumbocontext organization, and the product site is jumbocontext.com.