marvis
EN IT
★ GitHub Become a beta tester
Marvis How it works

Your agent is brilliant. Give it a brain.

Marvis is the brain you and your agents share. It captures the decisions behind your work, connects them across every project, and gets sharper over time, so nothing important slips and you always know what depends on what.

A coding agent that forgets between sessions A friendly agent. The memory floating above its head empties out, and a row of sessions shows that it starts from zero every time. context your agent session 1 session 2 session 3 starts from zero every time the agent today
00 The problem

Your agent is brilliant. And it works in the dark.

Every session starts from scratch. It can't see the decisions you made last week, or how your projects connect. So it re-guesses, and the work quietly drifts.

The more you build with it, the more piles up: choices nobody wrote down, context that lives only in your head, projects that should talk to each other but don't.

One command installs a memory Marvis connects to your agent. The agent's screen lights up with a small memory indicator and a green check, meaning it now remembers. Below, the install command. marvis now it shares your brain $uv tool install marvisx-cli✓ ready one line, in the terminal you already use
01 Install

One line, and it gets a brain.

You give your agent a single instruction. It sets up Marvis, the brain you'll share, right where you already work.

No dashboard to learn, no account to wire up first. From now on it works with the full picture, not a blank slate.

Marvis reads the work you already have Three project cards on the left feed into the brain on the right, with Marvis above it. Existing repos, notes and decisions are pulled in so the memory starts full. You can also start a fresh project. your projects brain into the brain existing work, or a fresh start
02 It reads your work

It learns the work you already have.

Marvis goes through the projects on your machine and pulls in the decisions, the reasons, and how things fit together. Your brain starts full, not empty.

Nothing to migrate by hand. Starting something new instead? Marvis grows that project's brain from day one.

You work in the same terminal as before A terminal window with a blinking cursor sits in front. Marvis watches quietly from the corner with a soft listening pulse. Your workflow does not change. marvis listens $claude › working on the feature… › same terminal, same flow nothing about your day changes
03 You keep working

You work exactly like before. But better.

Same terminal, same agent, same habits. As you work, Marvis captures the decisions, and the why behind them, the part that usually lives only in your head.

It doesn't ask you to fill in forms or change your routine. It quietly turns your everyday work into shared knowledge.

It remembers and connects the dots across projects A friendly constellation. A central memory connects to a decision, a project, a file and a note with soft lines that draw in one by one. brain a decision a project a file a note what links to what, kept across projects
04 It connects

It connects the dots across projects.

Every decision, project, file and note gets linked. Ask why something was done, or what relates to what, and the thread is right there, across all your work.

Search and the map of how things connect run locally, on your machine. The deeper thinking uses your own AI key, the same one your agent already uses.

One change, traced to everywhere it matters You change the VAT rate. Marvis lights up arrows to everywhere it lands across your projects: the service page, the quote spreadsheet, and the invoice and terms. you change one thing VAT rate service page quote sheet invoice + terms one change, everywhere it lands
05 What you can do now

Change one thing. See everywhere it matters.

Because the brain connects your projects, your agent can finally answer the question it never could before: if I change this, what else moves?

Say the VAT rate changes. Marvis points you straight to what it touches: the pricing on your service page, the spreadsheet that calculates your quotes, the invoice and the terms you already sent. One change, and nothing slips through.

It's yours. And it keeps getting sharper.

Marvis is local-first, so the brain lives on your own machine, not someone else's cloud. It learns from what goes wrong, turns repeated fixes into reusable know-how, and stays under your rules every step. Tell your agent, or install the command line.

in your agent › install Marvis, then run marvis init
or the cli $ uv tool install marvisx-cli
★ Star on GitHub → Become a beta tester
FAQ

Honest answers

A CLI sounds like technical stuff. I don't really know what that is.
You don't have to. If you already work with Claude Code, Codex or Cursor, your agent runs the commands for you: tell it to install Marvis and set it up. There's nothing new to learn on your end, and if you do want to learn how it works, just ask your agent.
My agent already keeps a CLAUDE.md or a context file. Isn't this the same?
Those store text. The difference is retrieval. Marvis maps how your code and decisions connect, so you can ask 'everywhere this is used' or 'what does changing this touch' and get an exact list, not a manual search through your files. For one tiny repo it isn't worth it; across several projects it is.
It's for people who write code. My work isn't really code.
Most people running it use it for non-code projects. The memory, the linking and the governance don't need source code at all; the code graph is just an extra layer when you do have a repo.
My agent has a huge context window. Do I really need Marvis?
The window resets every session. What you actually pay for is re-explaining a decision from three weeks ago, or not noticing two projects disagree. That gap between sessions is the whole point.
Is it solid enough to rely on?
It reads the projects you already have without touching a single byte of them, maps how your code and decisions connect, answers 'what does changing this affect', and installs guardrails your agent can't quietly break, all of it reversible. It's young and moving fast, but the parts you'd actually lean on are the ones that already work.
Can I trust the search, or is local too weak?
The exact answers don't rely on guesswork: keyword and the graph are deterministic and local, 'where is this defined', 'what calls it', 'what breaks if I change it'. Meaning-based search runs locally too, and the deeper reasoning uses your own model key, the same one your agent already uses.
I already use mem0 or Letta (or OpenClaw, Hermes). Why add Marvis?
Most of those are memory you build into your own app, or a store you query for recall. Marvis isn't a library to build with: your existing agent installs it and uses it as it is. And it adds two things a plain memory layer doesn't have: a deterministic map of how your code actually connects, so 'what does changing this affect' is an exact answer rather than a similarity guess, plus governance and an audit trail. If what you already run covers that, keep it.
Is it safe to point at all my projects?
Everything stays in a local database on your machine; the only thing that leaves is your own model calls. It's open-core, it doesn't touch your source files, and running git status right after setup shows exactly what it changed.
I'm not sure it's worth the setup.
You can't really tell whether a memory layer earns its place by reasoning about it. It's free, it runs locally, and you can remove it, so install it on a project you already have, use it for a week, and keep it only if you re-explain less and lose less. The cost of trying is near zero.