An Elephant Never Forgets
Neither Should

Jumbo is memory and context orchestration for coding agents.

Can we address the elephant in the room?

Working with coding agents is amazing. But let's be honest, it's not without frustration. Here are the common issues:

Agent Amnesia — Every session that came before is forgotten. You spend time and energy getting the agent caught up before you even start thinking about your goals.

Slop — AI without guardrails can produce code that kind of works, but it's rarely production-ready. That's not a productivity boost — it's a hassle.

Vendor lock-in — If an agent harness holds your memory, switching tools means losing that context. Your context should move with you.

How does Jumbo help?

Jumbo was created to address these frustrations and unlock the full joy of coding with agents.

Memory

Details about your project are saved, so your agents always have the correct context. Agent Amnesia is forgotten.

Quality

Your coding agents write shippable code on the first shot. You save time and tokens.

Orchestration

Run different agents in parallel. Optimize for capability and cost. Jumbo keeps everything in sync.

Interoperable

Jumbo is harness and model agnostic. Use them interchangeably or in parallel.

Portability

Switch to new models when they're released. Your context stays with you.

Those solve the core problems. These make Jumbo pleasant to use:

Context windows

Run agents longer without context rot.

Automatic

Hooks into your agent session and orchestrates the flow. It just works.

Full control

Jumbo's memories are yours. Manage your data directly from the terminal.

Private

All data stays local. Nothing leaves your machine.

Fast

No network calls. No lag. Everything runs locally.

"Jumbo is genuinely fascinating because it solves a core problem most builders are hitting right now: coordinating agents around a single goal without everything falling apart. No endless new threads when the AI loses context. You just build - and complete."

— Lorenz Flechtenmacher, CEO dewa

How does it work?

You drive. Jumbo orchestrates. Your agents execute.

You

Define

Set an objective, success criteria, and scope. Jumbo helps you break big tasks into right-sized goals.

1

Refine

Your agent couples relevant memories to the goal and builds a curated context packet.

2

Implement

Your agent receives curated context and gets to work. You collaborate, steer, or let it run autonomously.

3

Review

Your agent reviews the work against your criteria and project spec. You approve, or it goes back for another pass.

4

Codify

New learnings are captured — decisions, patterns, constraints. Your project memory grows with every goal.

You don't have to remember all these steps. Jumbo hooks into your agent sessions and guides the entire flow.

Plays Well with All Harnesses and Models

Jumbo works with any coding agent harness that supports AGENTS.md and skills built on the open agent skill standard.

Jumbo

Get Started

Install and run Jumbo.

$ npm i -g jumbo-cli
$ jumbo |

Jumbo will guide you on getting started.

FAQs

Jumbo is a CLI tool that gives your coding agents persistent memory and structured project context, turning them from makers of workable prototypes into builders of production-quality software.

Exactly! Jumbo is essentially context engineering with a fraction of the effort required to maintain markdown files and with better results. It works by collecting insights about your project as they arise in your agent workflow and serving context automatically to your coding agent.

Through hooks, with fallback to AGENTS.md. Your agent calls jumbo session start at the beginning of a session, and Jumbo injects relevant project context. A richer context packet is delivered when the agent starts work on a goal. New insights are captured in the natural flow of your agent conversations.

Change agents and models at will. Jumbo is harness- and model-agnostic. It works with any coding agent that supports AGENTS.md and truly excels with harnesses that support open agent skills. Jumbo just picks up where you left off.

Absolutely. You control how you want your agent to interact with Jumbo. Stay in-the-loop by approving each command, or run with pre-approved Jumbo commands for an automated experience.

Not at all. Jumbo prescribes an opinionated workflow that you can always bypass. It works alongside your agent to enhance its capabilities.

All data stays local. Nothing leaves your machine. Jumbo stores everything in .jumbo/ in your project directory.

Jumbo goes beyond static markdown files. It's an immutable event stream — capturing your entire project history, always current and auditable. You stay in your flow, never repeat yourself — only add new information when you need to. Markdown is a snapshot in time, Jumbo is your project's living memory.

Three commands: npm i -g jumbo-cli, then jumbo init in your project, then start your coding agent. Jumbo hooks in automatically.

Jumbo is currently optimized for single developer use. A cloud version for teams — Jumbo Herd — is under development. Sign up to get notified when it launches.

Cross my heart and hope to die (a.k.a. Promises)

Jumbo is free and open source for single developer use. Forever.