We started this to learn something. It became an IDE.
Two years ago this was a curiosity project — an attempt to understand how AI coding tools actually work under the hood. One question changed its direction, and it has not stopped growing since.
Three questions, two years, one editor
Nothing here was planned on a whiteboard. Each stage began as a question we could not stop poking at.
2024 — “How does any of this actually work?”
We started building to learn, not to ship. Completions, embeddings, retrieval, tool calling — we wanted to understand the machinery instead of trusting a demo video. Every answer we found opened a more interesting problem behind it.
The turn — “Why can’t we build an agentic one?”
That question became the motto. Not autocomplete that guesses the next line, but something that could read a repository, form a plan, edit across files, run the command, and show its work. Answering it properly meant the toy had to become real software.
Today — “Why is this a plugin?”
An agent bolted onto an editor is always a guest in someone else’s house. So we stopped decorating and rebuilt the house: an editor where the AI reaches every surface — the tree, the diff, the terminal, the index, the browser.
The experiment stopped being an experiment
Somewhere between the second and third question, what we had been building for ourselves turned into the editor we actually wanted to open every morning. That is a very different bar from “a fun weekend project”, and it changed how we worked: ship it, use it all day, find where it breaks, fix that first.
We still work exactly that way. Most of the roadmap is a list of things that annoyed us last week — which is why the fixes tend to land in the places you actually live in.
Built in Bengaluru
Made in Bengaluru, India, by developers with the same problems as the people who use it. We are early and the product is in beta — but the direction has been settled for a while now, and every release moves further along it.
The convenience of the cloud, on hardware you already own
Modern development leans harder on remote infrastructure every year. We wanted to find out how much of it could come home.
The default path
Your code is uploaded to be understood. Your context lives on someone else’s server. Your workflow depends on a subscription, a connection, and a rate limit — and the bill grows with your curiosity.
The path we’re building
Your machine indexes your codebase. Models run on your own silicon. Requests go straight from your computer to the provider you chose — or nowhere at all. The power is already under your desk; the software just has to use it.
What two years of that question produced
The whole product, in one place. Every item below ships today — nothing here is a roadmap promise.
EditVolt Chat — six modes, one panel
Ask for read-only answers. Plan to think a change through before a byte moves. Debug gathers evidence before it acts. Agent does the work. Design drives a live in-editor browser it can actually see. Multitask runs several workers in parallel. Ask and Plan never touch your files until you approve.
Pull in exactly the context you want with @-mentions — files, folders, symbols, problems, git changes, terminals, docs and URLs — and paste screenshots or attach images, with vision working across every connected provider.
EditVolt Complete — completions at 30,000 feet
Inline completions run on a bundled local engine: no cloud round-trip, no API key, and no internet required. Streaming multi-line suggestions arrive as you type, change propagation carries an edit across related code, and context comes from nearby code and your recent edits.
Tab-to-jump predicts where your next edit belongs and takes you there. Cloud completions exist too — as an explicit opt-in, never a default.
Agent mode — autonomy with a paper trail
The agent reads, writes, and runs commands across a full tool surface, and full-file writes stream directly into your open editor. Every edit is staged as a reviewable diff — nothing is ever auto-applied — and checkpoints roll the workspace back to any point in the conversation.
“Review My Changes” gives you one unified multi-file diff. Command consent is yours to set: auto-review, an allowlist, or run-everything. File-deletion protection is on by default. With no folder open, the agent’s first edit scaffolds a managed project in place.
Local codebase intelligence
EditVolt builds a vector index of your repository using an embedding model that ships inside the app and never touches the network. The database lives on disk, per workspace, and indexing works entirely offline.
Hybrid retrieval combines semantic search with fast text search to power @Workspace answers and the agent’s own code discovery. An optional local memory can learn your project over time — off by default, because it should be your choice.
On-device models, in the box
Chat and agent work on a fresh install with no API key and no account, because the models ship with the app: Gemma 4 in E2B, E4B and 12B tiers, Nemotron Nano 12B v2, and Qwen3.8 27B for machines with room to spare.
Pick a tier to trade quality against speed and footprint, or point EditVolt at Ollama and run anything else you like — still key-free, still local.
Bring any model — direct, no middleman
Anthropic Claude with your API key, or by bridging the Claude Code CLI sign-in you already have. Google Gemini with an AI Studio key. Any OpenAI-compatible endpoint — OpenAI, OpenRouter, Azure, vLLM, LM Studio. Hugging Face, remote or local. Ollama for fully local inference.
Requests go straight from your machine to the provider. We run no proxy and no backend in between, so AI usage is billed by your provider at their rates and never by us. Model lists update live, and smart routing can send quick work to a fast lane and heavy work to your frontier model.
A real customization substrate
Shape how the AI behaves in your repository with plain files in .editvolt/ —
versionable, reviewable, and shareable with your team. Rules packs get injected into every
conversation. Skills and subagents capture repeatable workflows like a security review or a docs
pass. Slash commands cover /explain, /fix, /test,
/review, /architect, /why, /debug and
/delegate.
Pre- and post-tool-use hooks live in hooks.json, MCP servers connect with per-tool
consent and allowlists, and instruction files you already keep — AGENTS.md, CLAUDE.md,
.cursorrules, copilot-instructions.md — are picked up as they are.
Privacy that is structural, not a setting
Telemetry isn’t switched off, it is removed from the binary — there is no event pipeline to re-enable. Strict privacy mode keeps every request on your machine. An ignore file walls off sensitive paths from the index and the agent. API keys live in your OS keychain, and updates are delivered over pinned certificates.
Signing in is entirely optional, unrelated to the AI features, and tells us only your email address and basic profile — never your code, which we could not see even if we wanted to.
Also in the box
Terminal and inline refine
Describe what you want in the terminal and get the command back. Refine a selection inline anywhere in the editor, without opening a panel.
AI commit messages
One click in the source-control panel writes a commit message from your staged diff — locally, if you want.
Project scaffolding
Start from an empty window. The agent scaffolds static sites and Next.js apps deterministically, then keeps working in place.
In-editor browser
An opt-in browser surface so agents can see what they build. Off by default, on when you want Design mode to have eyes.
VS Code under the hood
Built on Code-OSS, so your keybindings, themes, and Open VSX extensions all work from day one. Nothing to relearn.
Updates that install themselves
New builds arrive in the background and apply on restart — Developer ID signed and notarized by Apple, every release.
Principles we build against
Not slogans — each of these is a decision already made in the product, and each one can be checked.
Local-first development
Your machine should be powerful enough to be your personal development cloud. Chat, the agent, completions, and codebase search all run on-device — no key and no account required.
Developer ownership
Your code, environment, and workflows stay yours. Bring your own provider keys and your requests go directly to that provider — we operate no proxy or backend in between.
Privacy by design
Not everything has to leave your computer to be useful. Telemetry isn’t switched off, it’s structurally removed from the binary — and strict privacy mode keeps every request local.
Less dependency
Building well shouldn’t require expensive infrastructure or a permanent connection. EditVolt keeps working on a plane, on bad hotel Wi-Fi, and on the day an API has an outage.
Powerful, but approachable
Advanced capability shouldn’t demand a configuration marathon. Sensible defaults on day one, with rules, skills, and hooks waiting for the day you want them.
By developers, for developers
We have the same problems our users do, because we use what we ship. The roadmap is mostly a list of things that annoyed us last week.
$0. Forever.
Not a trial, not a free tier with something better behind it. There is nothing else to buy.
The whole editor, free
Every feature on this page, unlimited projects, unlimited chat, completions and agent tasks. No seats, no quotas, no card.
The same deal for everyone
Students, maintainers and companies all get the identical editor. There is nothing to discount, because there is nothing to charge for.
You pay your provider, not us
Run the bundled on-device models and pay nothing at all. Prefer a frontier model? Bring your own key and your provider bills you directly, at their rates.
Bigger than building another IDE
Rethinking what a developer’s computer can be
We want the machine you already own to act as workspace, development environment, compute platform, and personal cloud — one place where your tools, your context, and your code live together instead of being scattered across a dozen services.
We’re not claiming to have finished that. We’re early, the product is in beta, and there is a great deal left to get right. But the direction is settled, and every release moves further toward it.
Where we are
Still learning, still experimenting, still building. What began in Bengaluru as a way to learn something new is becoming our attempt to build something genuinely useful for developers everywhere.
Your machine. Your code. Your cloud.
Free forever, in beta, and ready to try today.