Built for developing with AI, not managing it

Choose the project.Start the agent.

If you move between Codex, Claude Code, and OpenCode, starting a session should not mean remembering folders, commands, accounts, and versions. Jota AI Launcher brings that starting point together with a finish desk for deciding what to resume, what to ship, and what to let go.

Download the app Why I built it

Open source
No telemetry
No shared keys
Windows + macOS

Local route / ready

Active project~/projects/my-application

AgentCodexREADY ↗

AgentClaude CodeREADY ↗

AgentOpenCodeREADY ↗

● Visible terminal. The agent opens inside the project you just selected.

The real application, not a concept

Jota AI Launcher · Local panel

Jota AI Launcher home screen with Codex, Claude Code, and OpenCode

One starting point. Three agents that keep using their own accounts and permissions.Jota AI Launcher

The story behind it

I built it because every session started before the work did.

I would open PowerShell, find the right folder, check which account was connected, try to remember how to resume a session, and see whether the CLI was current. None of those jobs was difficult. Together, they were noise.

When you develop with AI, the valuable work is holding onto the product's intent and deciding what should happen next. Every window switch and memorized command takes away a small piece of that attention. I wanted it back.

So I did not build another assistant that promised to think for me. I built a small, honest tool that helps me get to the right place, with the right agent, and begin.

Benefits you notice while working

Less ritual. More useful context.

It does not try to replace the way you develop. It organizes the starting point so your attention can stay with the code, the product, and the decisions that matter.

PATH

The right project is always visible

Opening an agent in the wrong folder costs more than a few seconds: it changes the context the agent receives and may lead it to edit the wrong thing. The launcher makes that choice explicit before you begin.

FLOW

Switch agents without rebuilding the ritual

Codex, Claude Code, and OpenCode have different strengths and workflows. A common starting point lets you choose for the task, not for the friction of opening each tool.

LOCAL

Your accounts remain yours

The launcher is not a middleman and has no API-key vault. Each CLI keeps its own session; someone else who installs the app must connect their own credentials.

STATE

See what is actually installed

Versions, account state, plugins, skills, and MCP servers appear in one inventory. Knowing your environment prevents blind troubleshooting and unnecessary updates.

New in version 0.5.0

Do not just open projects. Finish them.

When projects multiply, the problem is no longer finding another idea. It is recovering context, protecting attention, and turning open work into one concrete next move.

Search and sort

Find the project that matters now

Fuzzy search tolerates incomplete names. Combine it with status filters and sorting to locate a project quickly, even when you cannot remember its exact name.

Focus projects

Three priorities, not thirty mental tabs

Keep up to three priority projects visible. The limit is intentional: it protects your attention without hiding the rest of your library.

Finish plan

Define what “done” means

Save the phase, a target date, the next action, and a verifiable definition of done. The plan fits one session and prevents more features from appearing without a clear exit.

Checkpoint and exit

Your future self will know how to continue

Close a session by recording what moved, what is blocked, and the first action under 45 minutes. If an idea no longer deserves work, close it intentionally and preserve what you learned.

FindFocusMoveFinish

How it works

Three decisions. No hidden magic.

Everything happens in front of you. The launcher organizes; the terminal and each agent remain the tools you already know.

Choose a project

Select a folder yourself or pick a card from the library of projects detected on your computer.

Check the environment

See which agents are installed, their version, account status, and the extensions available to them.

Open the agent

Choose start or resume. A visible terminal opens in the selected folder, ready for work.

Jota AI Launcher's visual project library

Local library

Your projects stop being paths you have to memorize.

Each card summarizes purpose from the README, stack, GitHub, and services such as Vercel, Netlify, or Render. It also includes AI-built apps, plugins, and work that exists only as a local folder.

Who it is for

For people who build, not for a label.

Vibe coder, developer, designer who codes, independent founder, or someone still learning: the shared point is using AI to turn an idea into something that works.

It may help you if…

you move between two or more agents depending on the task;
you keep several local projects and want to recognize them at a glance;
you accumulate started projects and need to decide what to resume or close;
you need to check versions, skills, or plugins without touring every CLI;
you want to share a tool without sharing your accounts.

What it does not pretend to be

it is not an AI model and does not decide which agent is “best”;
it does not read the content of your conversations;
it does not hide installation or update commands;
it does not replace each agent's own permissions and confirmations.

Verifiable trust

You do not have to trust a promise.

A desktop application deserves more explanation than a download button. That is why the source, builds, and checks are public.

OpenPublic source codeYou can review exactly what the launcher does and how it opens each tool.

LocalNo backend or telemetryYour tool inventory and project paths are not sent to Jota.

ReproducibleGitHub Actions buildsWindows and macOS packages are produced through visible workflows.

VerifiableSHA-256 + attestationEach release includes checksums and provenance tied to its commit.

InventoriedDependency SBOMThe CycloneDX list shows which components are present in each version.

AnalyzedCodeQL and npm auditChanges pass Windows, macOS, and security checks.

Read the security review ↗How to verify a download ↗Report a vulnerability ↗

No sales pitch. No registration.

If it helps you begin with more clarity, use it.

Jota AI Launcher is free and open source. Download it, inspect the code, adapt it, or simply keep the idea: good tools are allowed to be quiet.

Download ↘Explore GitHub ↗