{"slug": "kilo-desktop-is-generally-available", "title": "Kilo Desktop Is Generally Available", "summary": "Kilo has released Kilo Desktop as generally available, following its October beta, offering a single app for running multiple coding agents in parallel across repositories with built-in diff review, notebooks, terminals, a browser, git, conda environment management, and native access to more than 500 models, including ones running locally. The app assigns each task its own session with status labels (Working, Requires approval, Requires input, Unread), supports a customizable Workspace Board kanban, and gives sessions isolated git worktrees so parallel agents do not conflict over the same files. Kilo positions the release around the claim that writing code is no longer the bottleneck and keeping track of agent work is, with Plan, Code, Ask, and Debug modes separating planning from execution.", "body_md": "We are excited to announce that [Kilo Desktop](https://kilo.ai/install) is now generally available. It’s the fastest way to go from an idea to a merged pull request with Kilo, and it’s built for the way you actually work with agents: several at once, across more than one repository, with your eyes on every diff. It also brings notebooks for research and data science into the same app, so you can explore datasets, test hypotheses, and turn what you learn into better decisions.\n\nWe [introduced Kilo Desktop in October](https://blog.kilo.ai/p/desktop), and brought coding agents, model choice, and data science into one app. Thanks to everyone who tried the beta and shared feedback.\n\nCoding agents got good enough that the bottleneck moved. Writing the code is no longer the slow part. Keeping track of it is. You have one agent refactoring the API, another chasing a flaky test, a third sketching a new screen, and they’re spread across terminal tabs, editor windows, and browser tabs you have to keep in sync by hand. Which one needs approval? Which branch is that on? Did it actually run the tests?\n\nKilo Desktop puts all of it in one window: agent sessions, notebooks, diffs, terminals, a browser, and git, with built-in conda environment management for your Python and data science projects. You also get native access to more than [500 models,](https://kilo.ai/leaderboard) including ones running on your own machine. It’s designed around three things: running agents in parallel safely, reviewing faster, and getting from a finished change to a pull request without the busywork.\n\n## One app. Every agent you’re running.\n\n#### Sessions you can actually keep track of\n\nEvery task gets its own session, and the sidebar shows what each one is doing: **Working**, **Requires approval**, **Requires input**, or **Unread**. Jump straight to the next session that needs you with a keystroke, pin the ones you’re actively driving, and filter by status or pull request. Kilo Desktop can also notify you when an agent finishes or needs you, and only when you’re not already looking if you prefer, so you can step away from a long task without missing the moment it needs a decision.\n\n#### A board for everything in flight\n\nWhen the list gets long, switch to the **[Workspace Board](https://kilo.ai/docs/desktop/features/workspace-board)**. Name your own columns, drag sessions between them as work moves from idea to review to merged, and pop the board out into its own window so it sits on a second monitor while you work. It’s a kanban for your agents, built from the sessions you already have.\n\n#### Isolated worktrees for parallel work\n\nRunning two agents in one checkout is how you get two agents fighting over the same file. When you start a session, pick a branch, or give the session [its own git worktree](https://kilo.ai/docs/desktop/features/git) so it works in complete isolation. Run five experiments side by side, keep the one that works, and delete the rest. The worktree is cleaned up with the session.\n\n#### Workspaces that span your whole project\n\nReal products rarely live in one repository. A Kilo Desktop workspace is a named group of folders, so the agent can read your frontend, your API, and your shared packages together. It can check the API contract while it updates the client, without you copying files between windows.\n\n#### Plan, then build\n\n**Plan** reads your project and returns a step-by-step implementation plan, opened in its own pane, without touching a line. Approve it and **Code** takes over, splitting the plan into subtasks and running them in parallel with sub-agents. You can open any sub-agent in its own pane to watch it work. **Ask** explains unfamiliar code without changing anything, and **Debug** finds what broke and fixes it.\n\n## One app. Review built in.\n\n#### A layout for reviewing as you go\n\nChats, diffs, files, terminals, and web pages all open as tabs, and you arrange them into split groups however the task needs. Keep the **Changes** pane beside the chat to see each edit land. Run the test suite in a terminal next to it. Preview your dev server in the [built-in browser](https://kilo.ai/docs/desktop/features/browser), DevTools included. For wide setups, carousel mode lets you scroll across panes at the width you choose, and any pane can pop out onto a second monitor.\n\nOnce you find an arrangement that works, save it as a **layout preset**. Apply it to any session from the Layouts menu, or make it your default so every new session opens ready to review.\n\n#### From diff to pull request in one click\n\nWhen the change looks right, **Commit, push & PR** does all three from the session header. Kilo writes the commit message and the pull request description from the actual diff, and it asks before you push to your default branch. Once the PR is open, its status shows on the session in the sidebar, so you always know which sessions have shipped.\n\n#### Rewind when an agent goes sideways\n\nEvery turn in the chat is a checkpoint. If the agent takes a wrong turn, **Revert to here** rolls back the conversation and its file changes, and you can restore them if you change your mind. In long sessions, a conversation map along the edge of the chat lets you jump between turns, **Find** (Mod+F) searches the chat or plan for the exact line you remember, and /compact keeps the conversation inside the model’s context.\n\n#### Control without the babysitting\n\nChoose how much rope the agent gets. **Review first** asks before editing files or running commands. **High autonomy** edits freely inside the workspace but still asks before running commands. You can switch between manual and automatic approval right from the composer whenever the stakes change.\n\n## One app. Massive power.\n\n#### 500+ models, and a router that picks for you\n\nEvery major frontier lab, the leading open-weight models, and providers from every geography are in one picker. Favorite the ones you use most and switch mid-project without changing tools. If you’d rather not choose, **[Auto Efficient](https://blog.kilo.ai/p/auto-efficient)** checks your request against [KiloBench](https://kilo.ai/kilobench) (a coding agent benchmark for comparing AI models in end-to-end coding tasks) and sends it to the least expensive model proven capable of that kind of task.\n\n#### Bring your own providers\n\nAlready paying for a model elsewhere? Add any OpenAI- or Anthropic-compatible endpoint with your own key and custom headers, or connect [Amazon Bedrock](https://kilo.ai/docs/desktop/settings/ai#ai-providers) using credentials already on your machine. It all shows up in the same picker.\n\n#### Local models\n\n[Some code should never leave your machine.](https://www.anaconda.com/blog/nvidia-rtx-spark-kilo-desktop-local-ai) The built-in [Local Model Server](https://kilo.ai/docs/desktop/features/local-inference) runs open GGUF models on your own hardware behind an OpenAI-compatible endpoint. Browse the Anaconda model catalog to see which models will fit your hardware before you download, then tune each model’s context and sampling, and decide whether it gets Kilo’s full coding toolkit or runs lean for quick questions.\n\n#### Skills, MCP servers, and custom agents\n\nThe [Kilo Marketplace](https://kilo.ai/docs/customize/marketplace) carries skills, Model Context Protocol (MCP) servers, and custom agents. Install any of them for one workspace or for all of them, and sign in to MCP servers right from settings. MCP servers that ship interactive apps render right in the chat, and the agent can show you mockups and diagrams inline while you design.\n\n#### Keyboard-first, and yours\n\nEverything has a shortcut, and every shortcut can be remapped. Open the command palette, start a session, open a terminal, or jump to a tab without touching the mouse. Mention files with @, run commands with /, and drop in screenshots. Queue your next prompts while the agent is still working, then pause, reorder, or edit the queue before it runs, so changing your mind doesn’t mean interrupting the agent. When you need your full editor, open any file in VS Code, Cursor, Zed, or a JetBrains IDE in one click.\n\n#### Bring your data science workflows and notebooks, too\n\nKilo Desktop puts coding, notebooks, and data analysis in the same workspace. Agents edit [live Jupyter notebooks](https://kilo.ai/docs/desktop/features/notebooks) directly, while built-in [conda environment management](https://kilo.ai/docs/desktop/features/environments) and access to Anaconda’s vetted packages help you get the project running. Explore your data, test a hypothesis, and put what you learn to work.\n\n### Built for you. Ready for your team.\n\nFor [teams and enterprises](https://kilo.ai/teams), Kilo Desktop gives administrators control over which models people can use, which providers and geographies are allowed, what each agent is permitted to do, and which skills it can install. Everyone builds faster, inside the boundaries your team sets.\n\nLeading a team? Read [Help Your Team Code Faster and Explore Further with Kilo Desktop](https://www.anaconda.com/blog/kilo-desktop-team) on the Anaconda blog.\n\nOpen a project, start a few sessions, and ship something. Open a notebook, explore your dataset, and make data-driven decisions. We’d love to see what you build.\n\n## **[Download Kilo Desktop](https://kilo.ai/install)**\n\n[Download Kilo Desktop](https://kilo.ai/install)", "url": "https://wpnews.pro/news/kilo-desktop-is-generally-available", "canonical_source": "https://blog.kilo.ai/p/kilo-desktop-is-generally-available", "published_at": "2026-11-17 16:41:47+00:00", "updated_at": "2026-11-17 20:20:05.683167+00:00", "lang": "en", "topics": ["ai-agents", "ai-tools", "developer-tools", "ai-products"], "entities": ["Kilo", "Kilo Desktop", "Workspace Board"], "also_reported_by": [], "alternates": {"html": "https://wpnews.pro/news/kilo-desktop-is-generally-available", "markdown": "https://wpnews.pro/news/kilo-desktop-is-generally-available.md", "text": "https://wpnews.pro/news/kilo-desktop-is-generally-available.txt", "jsonld": "https://wpnews.pro/news/kilo-desktop-is-generally-available.jsonld"}}