Start with the real project.
Point Codevalanche at the folder where the work already lives. The agent gets context without moving the repository into a separate hosted workspace.
❯ open ~/Projects/your-app
Local-first by design
Agents.
One home.
Your agents, repositories, and review loop. Together in one local-first workspace.
Built by Radovar. Your machine. Your final say.
THE WORKSPACE01 — FROM PROMPT TO PROOF
Bring the agents you already run
Claude Code
Codex
Antigravity
Grok
Local models
More as the product grows
The useful path
Your project stays where the work already is. Everything else follows.
No upload step. No second copy of the repository. No mystery between an agent's suggestion and your approval.
The product
One workspace. Every part of the loop.
Codevalanche is a desktop control plane for agentic work. It keeps the objective, the environment, the live process, and the proof attached to the same piece of work.
RUNTIMES / YOUR CHOICE
Keep the tools you already trust.
Use the provider and model that fit the task while Codevalanche gives the work a consistent project, permission, and review surface.
First-class accounts
Two Claude plans? Run both.
Sign in with the coding agents you already pay for, run local models entirely on your machine, and connect more than one account per provider. When one plan hits its limit, the task continues on the next.
Subscriptions, not API bills
Claude Code on your Claude plan. Codex on its plan. No new API keys, no markup, just the agents you already pay for.
Local models, zero meter
Run models fully on your machine. No usage caps, no API bill, and your code never leaves the device.
×2
Two accounts, one provider
Connect personal and work Claude accounts side by side. Each keeps its own session and meter. When one hits its limit, continue on the other.
Claude personal hits its limit. The same task continues on Claude studio.
Parallel work
Several tasks. Zero tab chaos.
Each objective keeps its own conversation, runtime, files, and process state. You see what needs attention and safely ignore the rest.
Illustrative workspace. Statuses come from the live session, not from a refresh button.
From intent to evidence
Let agents move fast. Keep the final say.
The interface is built around the moments that matter: starting cleanly, seeing what changed, and knowing what actually happened.
Give the work a home.
Each objective keeps its conversation, runtime, files, and process state together, so parallel work does not dissolve into open tabs.
project contextlive processreview state
Inspect before you accept.
Changes arrive as visible work. Review the diff, check the result, and decide what belongs in your codebase.
src/queue/retry.ts +2 −1
- await wait(1000)
+ await wait(backoff(attempt))
+ attempt += 1
Leave with evidence.
Keep the commands, preview, checks, and delivery state close to the task that produced them.
✓ checks✓ preview✓ diffcommit
Considered down to the details
All the small things.
Less friction in the everyday. More room for the work that matters.
Projects that stay organized
Groups, projects, and tasks in one sidebar. Drag work where it belongs and find it there tomorrow.
Usage you can see
Plan windows and limits stay visible next to the work, before they surprise you.
Permissions at the boundary
Every sensitive action asks at the moment it happens, with the exact command in view.
A real browser, inside
Preview the running app next to the conversation that changed it. Evidence beats screenshots from memory.
History that follows you
Import existing agent conversations and keep them synced next to the projects they belong to.
Models that fit your machine
The local catalog checks your hardware before you download, not after.
Speaks your language
The interface ships in 17 languages, right-to-left included, and follows your system by default.
Englishالعربية الفصحىDeutschEspañolFrançaisहिन्दीBahasa IndonesiaItaliano日本語한국어Português (Brasil)RomânăРусскийTürkçeTiếng Việt简体中文繁體中文
Work in context
The surface changes. The thread stays intact.
A PROJECT HAS A MEMORY
Keep the conversation next to the code.
Projects, tasks, and workspace state remain visible together so you can return to the actual line of work instead of reconstructing context from scratch.
THE RUNTIME IS A CHOICE
Use the model that fits the moment.
Codevalanche gives the selected runtime room to work while keeping the surrounding project surface consistent and reviewable.
PROGRESS IS NOT PROOF
See what happened before you call it done.
Changes, checks, previews, and delivery state belong in the same loop, so a polished answer never has to stand in for evidence.
LOCAL BY DEFAULT
Put the boundary where the work is.
Your repository stays on your machine while the selected provider receives only the context required for its own run.
A focused interface for work that deserves a visible trail.
The companion
Step away. Stay in the loop.
Agents keep working after you stand up. The phone companion, now in development, pairs with your desktop, shows every run, and lets you unblock the one that needs an answer.
01
Watch the fleet
Working, needs you, ready to review. The same statuses as your desktop, in your pocket.
02
Approve from anywhere
Read the diff, then approve or decline. Execution and code never leave your machine.
03
Start the next task
Queue new work from the couch. It is waiting, reviewed and visible, when you sit back down.
Companion app in development, paired to the desktop workspace.
Local-first by design
Your repository is not a marketing asset.
Codevalanche is built around a simple boundary: keep the workspace layer close to the person doing the work, and make every provider handoff explicit.
01
Real files
Agents work against the project you opened, not a pasted approximation.
02
Explicit permissions
Actions are visible at the boundary where they happen, with you in the approval loop.
03
Portable providers
Keep the accounts, models, and tools you trust while Codevalanche organizes the work around them.
04
Reviewable results
Every meaningful change has a place to be inspected before it becomes part of the project.
Get Codevalanche
Built for the machine in front of you.
The workspace runs on your desktop. This page already knows which build belongs to you.
You are reading this on a phone.
Codevalanche is a desktop workspace, and the phone companion pairs with it. Open this page on your computer when you are back at the desk. The right build appears here the moment it ships.
Codevalanche for desktop
Pick your platform below
Detected from your browser: …. Wrong guess? Every build is listed below.
Windows
64-bit installer
In preparationmacOS
Apple Silicon
In preparationmacOS
Intel
In preparationLinux
x64
In preparationFirst public builds are being prepared. No waitlist, no email. This page simply updates when they land.
The standard
Good tooling should leave you calmer than it found you.
More context without more chaos. More capability without giving up the boundary.
Questions, answered
A clear product deserves a clear boundary.
What is Codevalanche?+
Codevalanche is a local-first desktop workspace that connects coding agents to real projects on your machine and keeps the work visible from prompt to review.
Does Codevalanche replace my coding agent?+
No. It provides the surrounding project, task, permission, process, and review layer while the selected coding-agent runtime keeps its own account and model behavior.
Can I connect more than one account of the same provider?+
Yes. Accounts are first-class in Codevalanche. You can keep, for example, a personal and a work Claude subscription connected at the same time, see each account's usage separately, and choose which one a task runs on.
Where does my repository live?+
The initial product is designed around the repository you already have on disk. It does not require an upload step to begin working.
Is Codevalanche available to download?+
The download section on this page lists every platform build and its current state. The first public builds are in preparation, and the buttons activate the moment they ship.