VD Agent

A desktop AI coding agent that works with free API keys.

VD Agent by Virender Dhiman · VD Agent License: free to use, credit required

What is VD Agent?

VD Agent is a desktop app that pairs an AI coding assistant with a code editor, terminal, and git client for the projects on your computer. You describe a task, and the agent reads your code, edits files, runs commands, and checks its own work with your project's tests and build.

It is built to work well on free API keys. It asks your key which models it can use, picks the strongest chat model, and moves on to the next one when a model is rate-limited or unavailable. It can also run fully offline with a local model such as Ollama.

Your chats, settings, and keys stay on your machine. There is no account and no telemetry.

Features

• Agent that verifies its work. Reads before it claims, pushes back on risky requests, runs your tests or build before it reports success, and asks before destructive actions.
• Modern agentic environments. Choose Normal, Smart, or Agentic. Substantial chat requests can move through a prompt analyst, optional specialists, tool-enabled worker, independent reviewer, bounded repair, and final verifier. Team presets, token budgets, custom roles, and a visual trace panel keep the workflow strong without wasting tokens. See Agentic Environments.
• Tools. Files, code search, shell commands, git, web search, and some media and Web3 helpers. See TOOLS.md.
• Safe archive-to-CSV workflows. Inventories ZIP files before extraction, rejects unsafe entries, and guides the agent to inspect every source, preserve provenance, and verify structured CSV output.
• Plan Mode. The agent proposes a plan and waits for your approval before it changes files.
• Automatic model selection. Discovers and ranks the models your key can call, and falls back between models and providers. See How model selection works.
• Durable resume and recovery. Chats, tasks, workspace state, recent tool outcomes, terminal context, and verification state survive restarts. Agent file edits get pre-change recovery checkpoints.
• Searchable chat history. Reopen, rename, pin, search, export, import, restore, or delete local sessions.
• Technical safety controls. Electron-native approval prompts protect commands and destructive Git/file actions; writes outside the open workspace are blocked.
• Built-in tools for developers. Monaco code editor, multi-tab terminal, git status, diff, commit, branches, and a command palette (Ctrl+P).
• Custom Rules. Your own conventions added to every request.

Download and Install

1. Open the Releases page and download the file for your system:

   System	File
   Windows 10/11 (64-bit)	VD Agent Setup <version>.exe (installer) or VD Agent <version>.exe (portable)
   macOS	VD Agent-<version>.dmg (Apple Silicon builds end in -arm64)
   Linux (64-bit)	VD Agent-<version>.AppImage or vd-agent_<version>_amd64.deb

2. Optional but recommended: check the file against SHA256SUMS.txt from the same release. See verify your download.

3. Install and launch it. If Windows SmartScreen or macOS Gatekeeper warns about an unsigned build, see first launch.

If no release is listed yet, run from source. Full instructions, updating, and uninstalling are in docs/INSTALL.md.

Run from source

You need Node.js 22.19 or newer, git, and native build tools for the terminal (node-pty). See prerequisites.

git clone https://github.com/virrdhiman/desktop-agent.git
cd desktop-agent
npm install
npm run dev

To build your own installer, run Win\build.bat, bash Mac/build.sh, or bash Linux/build.sh. See docs/RELEASE_CHECKLIST.md.

Free API key setup

You need one key to start. Groq and Google Gemini have free tiers and are good first choices.

1. Open Settings (⚙️ in the sidebar) and select Groq or Google Gemini.
2. Click Get API Key. The provider's site opens in your browser. Create a key there and copy it.
3. Paste the key into VD Agent and click Save. Selecting a provider in the list also makes it the active one.
4. Open a project folder (📁 Files → Open), go to 🤖 Agent, and send a message.

No key? Install Ollama, run ollama pull qwen2.5-coder:7b, and pick Ollama (Local) in Settings. Prompts never leave your machine.

Adding keys for more than one free provider makes rate limits less of a problem, because VD Agent can fall back between them. Free-tier limits are set by each provider and change often. See docs/PROVIDERS.md for other providers and their trade-offs.

How model selection works

For OpenAI-compatible providers, VD Agent:

1. Lists the models your key can call and drops non-chat models such as embeddings, speech, image, and moderation models.
2. Ranks the rest by coding quality and efficiency, then learns from local success, failure, and latency history. If a live list is available, stale saved model IDs are dropped automatically.
3. Moves to the next model on rate limits (429), model-not-found or unsupported model errors, overload, and 503s.
4. Stops immediately on an invalid key, so you can fix it instead of having the problem hidden.
5. Saves the refreshed live model list and the model that worked, unless you pinned a model in the Agent header.

If the active provider fails for any reason other than an invalid key, VD Agent tries other official free providers that have keys. Local servers are never used automatically. The catalog is official APIs and local OpenAI-compatible servers only.

Privacy and local data

VD Agent has no server, account, telemetry, or analytics. Everything it stores stays in your user data folder:

• Chats: conversations/<id>.json, one file per chat. API keys are redacted before a chat is written.
• Recovery checkpoints: checkpoints/<session-id>/, bounded pre-edit copies of files changed by agent file tools.
• Project memory: project-memory/, a deterministic local summary used to resume work with repository context.
• Settings: settings.json, with provider configuration, custom rules, and preferences.
• API keys: stored in settings.json and encrypted with your OS keychain (Electron safeStorage) when one is available. On Linux without a keyring they are stored unencrypted. Keys are never included in settings exports.
• Private diagnostics: an optional local JSON export contains app/runtime metadata and aggregate health counts only. It excludes API keys, chats, filenames, workspace paths, and source code.

Your prompts and code go only to the AI provider you choose, plus any service an agent tool calls on your behalf (for example web search). Details are in docs/PRIVACY.md. The agent runs commands with your user permissions, so read docs/SECURITY.md before pointing it at important work.

Documentation

Support the project

If VD Agent saves you time, starring the GitHub repository helps other people find it. Starring is a request, not a condition of the license.

If you share, review, demo, or write about VD Agent, the license asks you to credit it:

VD Agent by Virender Dhiman - https://virender.in

Please link to https://github.com/virrdhiman/desktop-agent as well. Bug reports and pull requests are welcome; see docs/CONTRIBUTING.md.

License

Copyright (c) 2025-2026 Virender Dhiman. All rights reserved.

VD Agent is proprietary, source-available software and the exclusive property of Virender Dhiman. From version 1.0.1 it is licensed under the VD Agent License:

Credit is required when you share or showcase it. VD Agent is built on open-source dependencies such as Electron, React, and Monaco Editor, which keep their own licenses (many are MIT-licensed); those licenses cover the dependencies, not VD Agent itself. This summary is for convenience and is not legal advice; the LICENSE file is the binding text. For permissions the license does not cover, contact the author through virender.in.