Chat with real context
Attach files, folders, your selection, problems or git changes with @, or paste screenshots and UI mockups for the model to follow. Large content is compressed to fit the model.
AI coding assistant for VS Code
Kingcurve brings chat, multi-file edits and a coding agent into VS Code. Run local models with Ollama or LM Studio, or bring your own OpenAI, Claude, Gemini or OpenRouter key. Nothing changes until you say so.
Free · MIT License · Requires VS Code 1.106 or newer
Works with
A complete AI workflow inside VS Code, built so that you stay in control of your code.
Attach files, folders, your selection, problems or git changes with @, or paste screenshots and UI mockups for the model to follow. Large content is compressed to fit the model.
Every change appears as a diff. Keep or drop individual files and changes, apply when ready, and roll back at any time.
Agent mode reads code, edits, runs commands and tests, and can hand parts of a task to team members on other models. Edits and commands wait for your approval.
Connect Model Context Protocol servers over stdio or HTTP, with allow, ask or deny for every tool.
Conversations are saved on your computer. See tokens and estimated cost by day, model and feature.
Generate commit messages from staged changes, review uncommitted work, and turn on inline completions if you want them.
Add Kingcurve from the Visual Studio Marketplace. It opens in the side bar on the right, next to your other AI chats.
Ollama and LM Studio are detected automatically. For cloud models, paste your API key.
Ask, Edit, Plan or Agent, depending on how much you want the AI to do.
Your code, your models, your decisions. Kingcurve never sits between you and your provider.
Yes. Kingcurve is free and released under the MIT License. Cloud providers bill you for your own API usage; local models through Ollama or LM Studio cost nothing extra.
Only to the model provider you select. Kingcurve has no server of its own and collects no telemetry. With a local model, everything stays on your machine.
Any model served by Ollama, LM Studio or an OpenAI-compatible server, plus OpenAI, Anthropic Claude, Google Gemini and OpenRouter with your own API key. Agent mode needs a model that supports tool calling.
Not by default. Every edit is shown as a diff and every command asks for approval. You can relax these rules in Settings, for example to run commands classified as safe.
Yes, with a local model. Kingcurve has no server of its own, so with Ollama or LM Studio running on your computer, chat, edits and the agent work without an internet connection.
Yes. Paste, drop or attach PNG, JPEG, WebP or GIF images (and text or code files), then describe what you want. This needs a model with vision, such as GPT-4o, Claude, Gemini, or a local vision model in Ollama or LM Studio.
Yes. In the Agent team tab you can add team members, such as Frontend, Tests or Docs. Each has its own model, allowed tools and the files it may change. In Agent mode the lead agent hands parts of a task to them, and every change still needs your approval. You can also choose a Planner model for Plan mode and a Reviewer model that checks changes.
Yes. Kingcurve follows the project instructions in AGENTS.md, the file most AI coding tools read, plus .kingcurve/rules and .kingcurve/memory.md. "Write Handoff" saves the state of a task to .kingcurve/handoff.md, so another agent or teammate can continue it. You can also save conversations as Markdown in .kingcurve/sessions and share an agent team through .kingcurve/team.json.
Visual Studio Code 1.106 or newer. Kingcurve is published on the Visual Studio Marketplace.
Free and MIT-licensed, on the Visual Studio Marketplace.