Nova AI brings the Nova model catalog into VS Code. Chat with an AI agent that reads your project, edits files, runs commands and searches the web, and stays in your control: every change can be reviewed, kept or undone.
Requirements
Getting started
- Install the Nova AI extension.
- Open the Nova AI view from the Activity Bar.
- Select Connect Nova AI and paste your API key. It is stored in VS Code Secret Storage, never in files or settings.
- Start typing in the Nova chat. You can also pick a Nova model in VS Code Chat, or type
@nova.
Stuck or curious? Run Nova AI: Help for a full guide inside VS Code.
Three ways to use Nova
| Where |
What it is |
| Nova chat (sidebar or editor tab) |
Nova's own agent: built-in tools, approvals, diffs, history, memory and skills. The main experience. |
| VS Code Chat |
Nova models in VS Code's chat model picker and agent mode. |
@nova |
Nova's chat participant inside VS Code Chat, with /explain, /fix, /tests and /compact. |
What the Nova chat can do
Work on your code
- Read, search and edit: find files, check problems, edit and create files, run commands, fetch web pages and use MCP tools registered in VS Code.
- Review every change: edit cards show the changed lines. A strip above the input lists the files Nova touched, with Keep and Undo per file or for all.
- Stay in control: choose how often Nova asks first with the permissions menu under the input (see Modes).
- Search the web: Nova looks up current information and cites its sources. No API key needed.
- Use your browser: with the Nova AI Browser integration Nova can open pages and take screenshots in Chrome, Edge or Brave. Set it up with Nova AI: Set Up Browser Integration (it runs automatically on startup by default).
- Research in parallel: for broad questions Nova can start up to a few research sub-agents. A strip above the input shows each one live; click an agent to watch what it does, or stop it.
- Plan multi-step work: Nova keeps a task list above the input and asks you questions with options when it needs a decision.
Modes and permissions
Two small menus under the input box:
| Agent mode |
For |
| Interactive |
Normal back-and-forth work (default) |
| Plan |
Researches and writes a plan; read-only, changes nothing |
| Ask |
Answers questions about your code; read-only |
| Autopilot |
Works on its own until the task is done, within your permissions |
Press Shift+Tab to cycle through the agents. Create your own with Nova AI: Manage Agents.
| Permissions |
Nova asks before |
| Manual |
every edit and command |
| Assisted (default) |
anything that changes something, unless a rule allows it |
| Allow all |
nothing |
On an approval card choose Allow, Always allow (saves a rule), Allow all for the chat, or Reject.
Give it context
- Type
@ to mention a workspace file.
- Use the paperclip, or paste and drop files and images (also from the Explorer; hold Shift while dropping from VS Code). Click an image to enlarge it.
- Right-click in the editor and choose Add to Nova Chat to send your selection.
- The ring next to Send shows how full the model's context window is; click it for details and a Compact button. Long chats are compacted automatically.
Chat comfortably
- Enter sends, Shift+Enter adds a line. Keep typing while Nova works to steer it, or queue a follow-up.
- Edit and resend an earlier message, or regenerate an answer and keep both versions.
- Replies render Markdown, code, links (checked before they open) and LaTeX math, with copy buttons.
- Thinking and tool calls fold into short, readable steps.
- Type
/ for quick commands: /clear, /compact, /model, /rename, /help.
- Chats (history button) lists every chat of the workspace with search, rename and delete. Open Chat in Editor moves a chat to its own tab, so several can run side by side.
Make Nova yours
Memory: Nova remembers your preferences (global) and notes about the project, and reads NOVA.md or AGENTS.md in your repository for team instructions. It asks before changing memory.
Skills: reusable instructions (a folder with SKILL.md) that Nova loads when a task fits. Manage them with Nova AI: Manage Skills, globally or per project.
Agents: custom agent modes as markdown files, global or per project.
Permission rules: shared or personal allow/deny rules, in .nova-ai/settings.json and .nova-ai/settings.local.json:
{ "permissions": { "allow": ["run_command(npm test*)", "edit_file"], "deny": ["run_command(rm -rf*)"] } }
A rule is a tool name, optionally with a * pattern. Deny always wins. Rules from a repository apply only in a trusted workspace.
Commands
Open the Command Palette (Ctrl+Shift+P / Cmd+Shift+P) and type Nova AI:
| Command |
What it does |
| Open Nova Chat / New Chat / Open Chat in Editor |
Show, start or move a chat |
| Chats / Search Chats |
Find, open, rename or delete chats |
| Add to Nova Chat |
Send the editor selection (also in the editor context menu) |
| Manage Skills / Manage Agents |
Create and manage skills and agent modes |
| Open Global Memory / Open Project Memory |
Edit Nova's memory |
| Set Up Browser Integration |
Let Nova use your browser |
| Account and Models / Manage Account / Refresh Models |
Connection and model list |
| Sign In / Sign Out |
Connect or disconnect your API key |
| Reveal Nova Folder / Clean Up Projects |
Open ~/.nova-ai, remove data of deleted workspaces |
| Manage Language Models / Open Chat / Open Settings |
VS Code integration |
| Help |
The full in-editor guide |
Settings
Search for Nova in VS Code Settings. The ones you are most likely to want:
| Setting |
Default |
What it does |
nova.agent.approvalMode |
autoReadOnly |
When Nova asks before running tools (ask, autoReadOnly, autoAll) |
nova.agent.maxToolRounds |
25 |
Tool rounds per request before Nova pauses |
nova.agent.commandTimeoutSeconds |
120 |
Time limit for commands Nova runs |
nova.agent.includeMcpTools |
on |
Offer MCP tools to the Nova chat |
nova.agent.subagents.* |
on |
Research sub-agents: on/off, model, and limits on how many run |
nova.webSearch.* |
on |
Web search providers (DuckDuckGo, You.com, your own SearXNG) |
nova.browser.autoSetup |
on |
Set up the browser integration on startup |
nova.context.autoCompact |
on |
Compact long conversations automatically |
nova.context.compactThreshold |
0.8 |
Share of the context window at which @nova compacts |
nova.context.defaultContextWindow |
32768 |
Context window for models that do not advertise one |
nova.memory.enabled |
on |
Use and update Nova's memory |
nova.home |
~/.nova-ai |
Where memory, chats, skills and scratch files live |
Developer settings (nova.chat.rewriteAssistantIdentity, nova.models.editTools, nova.developer.parseModelCapabilities, nova.enableDiagnostics) are in the Developer Settings group.
Some integrations (thinking output, per-model options in the model picker) use VS Code API proposals. They work where VS Code enables them and are skipped otherwise.
Where your data lives
Nova keeps its data in ~/.nova-ai, readable only by you:
~/.nova-ai/
MEMORY.md global memory
skills/ agents/ your global skills and agents
settings.json your own permission rules
projects/<name>-<hash>/ per workspace: memory, chat history, scratch files
A repository can also hold a .nova-ai/ folder (rules, skills, agents) and a NOVA.md / AGENTS.md with team instructions. The nova-ai terminal app shares this folder, so memory, skills and chats carry over. The layout is documented in docs/NOVA_HOME.md.
Troubleshooting
- No models: check the connection in the Nova AI view, then run Nova AI: Refresh Models.
- "Stopped after 25 tool rounds": send
continue, or raise nova.agent.maxToolRounds.
- Conversation no longer fits: use
/compact, pick a model with a larger context window, or start over with /clear.
- Something else: turn on
nova.enableDiagnostics and read the Nova AI output channel.
Privacy & security
Your API key lives in VS Code Secret Storage. Nova data files are created readable only by you. Links in replies are never opened without showing you where they go.
AI can make mistakes. Review important output before relying on it.
For contributing and development details, see docs/DEVELOPMENT.md.
| |
