FAQ
Quick answers to the questions everyone asks
Quick answers to the questions everyone asks. Each answer links to the page with the full story.
Is my code sent anywhere besides my chosen AI provider?
Section titled “Is my code sent anywhere besides my chosen AI provider?”No. Nexgile Code has no cloud back-end of its own: requests go directly from your machine to the AI provider you configured, and nowhere else. API keys are stored in your editor’s secret storage on your machine. The telemetry client is a no-op — nothing is sent over the network. What the provider does with requests is governed by your agreement with them, which is exactly why the provider choice is yours. See Providers Overview.
Can I use it completely offline?
Section titled “Can I use it completely offline?”Yes — run a local model through Ollama or LM Studio (or any local OpenAI-compatible endpoint). Everything except the model call already runs locally. Semantic codebase indexing can also stay offline with a local Qdrant container and an Ollama embedding model.
What does it cost?
Section titled “What does it cost?”The product doesn’t meter you — you bring your own provider API key and pay the provider for tokens. The task header shows a live cost meter for the current task, and you can set hard caps: allowedMaxCost (USD per task) and allowedMaxRequests (LLM calls per task) pause the task at the limit with a “Reset and Continue” prompt. See Model Behavior & Costs.
Can it run commands or edit files without asking me?
Section titled “Can it run commands or edit files without asking me?”Only if you explicitly turn that on. By default every file edit shows a diff and every command shows the exact command line for your approval. Auto-approval is opt-in per category — and auto-approved commands are still filtered through your allowedCommands/deniedCommands prefix lists.
What’s the difference between modes, agents, skills, and rules?
Section titled “What’s the difference between modes, agents, skills, and rules?”- Modes are personas with tool permissions — they decide what the agent can do (Code edits and runs commands; Ask is read-only).
- Custom agents are your own mode definitions written one-per-file in
.nexgile/agents/, live-reloaded. - Skills are instruction packages the agent loads on demand for a specific procedure — loaded only when needed, so they cost no context until used.
- Rules (
.nexgilerules,AGENTS.md) are conventions injected into every prompt — always on.
Can I undo what the agent did?
Section titled “Can I undo what the agent did?”Yes — every agent-initiated change is captured in a shadow git repository as a checkpoint, with one-click restore and diffs between any two points. The one gap: checkpoints restore files, not terminal side effects — a npm install or a database migration isn’t reversed by a restore.
Does it work in my language?
Section titled “Does it work in my language?”Yes. Set your language preference in settings and the agent thinks and responds in it — 18 languages are supported: English, Català, Deutsch, Español, Français, हिन्दी, Bahasa Indonesia, Italiano, 日本語, 한국어, Nederlands, Polski, Português (BR), Русский, Türkçe, Tiếng Việt, 简体中文, 繁體中文. You can also just type in your language — the agent follows your lead.
How do I teach it our company’s coding conventions?
Section titled “How do I teach it our company’s coding conventions?”Put them in rules files: .nexgilerules at the project root is injected into every prompt (commit it so the whole team shares it), .nexgile/rules/ holds an organized directory of rules, and per-mode variants apply conventions only where they matter. Run the built-in /init command to have the agent analyze the codebase and generate an AGENTS.md starting point.
Can it access files outside my workspace?
Section titled “Can it access files outside my workspace?”Only with your approval — reads and writes outside the workspace are flagged distinctly in approval prompts, and auto-approving them requires the separate alwaysAllowReadOnlyOutsideWorkspace / alwaysAllowWriteOutsideWorkspace toggles, which are off by default. See Auto-Approving Actions.
How do I keep it away from secrets?
Section titled “How do I keep it away from secrets?”Create a .nexgileignore at the workspace root with gitignore syntax — .env, key material, customer data dumps, build artifacts. Ignored files are hidden from the agent’s reads, searches, and context. Belt-and-suspenders: real secrets shouldn’t live in the repo at all.
Which model should I pick?
Section titled “Which model should I pick?”A strong recent coding model from a major provider for Code/Debug/Workflow tasks, and a cheaper, faster model for Ask, documentation, and routine chores. Create one API configuration profile for each and pin them per mode — the providers overview compares what each provider is good at. When in doubt, start strong: a better model that solves the task in one pass is usually cheaper than a weak one that loops.
Can multiple tasks run in parallel?
Section titled “Can multiple tasks run in parallel?”One task is active per panel, and subtasks run sequentially — the parent pauses while a child works (in Workflow Mode, Tester and Reviewer also run one after the other). For genuinely parallel work, use worktrees: each git worktree opens in its own window with its own Nexgile Code panel and its own task.
Do I need Docker?
Section titled “Do I need Docker?”Only for semantic codebase search, which uses a Qdrant vector store (docker run -p 6333:6333 qdrant/qdrant). Every other feature works without Docker.
Can I use my ChatGPT subscription instead of an API key?
Section titled “Can I use my ChatGPT subscription instead of an API key?”Yes — the provider list includes OpenAI - ChatGPT Plus/Pro, which signs in with your ChatGPT subscription rather than a platform API key. See Providers Overview for the full roster of 21 providers.
Why did my task stop in the middle?
Section titled “Why did my task stop in the middle?”Three usual reasons: it’s waiting for an approval (check the chat for a pending card), it hit a request or cost cap (a “Reset and Continue” prompt appears), or a follow-up question is waiting for your answer. None of these lose work — answer, approve, or continue and the task resumes. See Troubleshooting.
How do I update Nx IDE?
Section titled “How do I update Nx IDE?”Nexgile Code comes built into Nx IDE — there’s nothing separate to install or update. Updates arrive through your Nexgile distribution channel; enterprise deployments receive packaged builds through their Nexgile representative. Your current version is in Settings → About.
Where is my data stored?
Section titled “Where is my data stored?”Locally. Task history lives in your editor’s global storage for the extension (tasks/<task-id>/ — the model conversation, the chat transcript, and metadata), global configuration in a settings/ folder next to it, and checkpoints in a shadow git repository outside your project. Set nexgile-code.customStoragePath to relocate storage. See Files & Directories.
Does it integrate with Jira or Linear?
Section titled “Does it integrate with Jira or Linear?”Not out of the box — Playwright is the only built-in MCP server. You connect Jira, Linear, or any other tracker yourself by adding it to your MCP configuration, after which its tools are available to the agent like any others. Setup for both the hosted-URL and local-package shapes is in Configuring MCP Servers.
One limit worth knowing before you plan around it: the five GitHub specialist modes drive the gh CLI and only understand GitHub. Handing 🔧 Issue Fixer a Jira key does nothing. With a tracker server connected, use 💻 Code, 🪲 Debug, or 🪃 Orchestrator and name the server in your request.
Can the agent fix an issue if I just give it the number?
Section titled “Can the agent fix an issue if I just give it the number?”For GitHub, yes. Give 🔧 Issue Fixer the issue URL (or 🕵 Issue Investigator, or 🛠 PR Fixer and 🔀 Merge Resolver a PR number) and it retrieves the issue with gh, reads the acceptance criteria, explores the affected files, and implements. This needs the gh CLI installed and authenticated — run gh auth status first.
For Jira or Linear, only once you’ve connected that tracker over MCP, and only from a general mode. The request looks like Use the jira MCP server to fetch PROJ-412 and implement it rather than a bare ticket key.
How do I reset settings?
Section titled “How do I reset settings?”Settings → About → Reset clears the in-app configuration (Export first if you might want it back; Import restores it). Project-level behavior is controlled by files in your repo — .nexgilerules, .nexgilemodes, .nexgile/ — which you manage through git like any other code.