# Nexrall Code **Your autonomous coding agent, right inside VS Code.** Nexrall Code is an AI agent that reads your codebase, edits files, runs commands, and ships work end-to-end — with you in control at every step. Point it at a bug, a feature, or a refactor, and it will explore the repo, make the changes, run the tests, and verify its own work. Powered by [Nexrall](https://nexrall.com), it routes your prompts to your choice of frontier models — Claude, GPT, DeepSeek and GLM — with up to **1M tokens of context**, through a single account instead of juggling API keys. --- ## How it works 1. **Describe the task** — tell Nexrall what you want in plain language (build, fix, refactor, explain), or right-click any code and choose *Ask About Selection*. 2. **Nexrall explores** — it maps symbols, reads relevant files, and searches the codebase to understand the project before touching anything. 3. **Nexrall acts** — with your permission, it edits files, runs builds and tests, and iterates until the job is done. 4. **You review** — every change appears as an inline diff in the chat. Accept it, tweak it, or rewind to any earlier checkpoint. ## Features ### 🌐 Any frontier model, up to 1M tokens of context Switch models from the chat menu without changing your workflow — Claude Sonnet 5.5, Opus 5.5 and Fable 5.1 (1M context each), GPT-6 Astra, GPT-6 Sol, GPT-6 Luna and GPT-5.6 Terra (922K), DeepSeek Flash, and GLM 5.3 (1M). Reasoning effort adapts to whichever provider you pick, so the slider always reflects what the model can actually do. ### ✏️ Edit files with inline diffs Every edit is shown as a diff you can expand right inside the chat — no tab-hopping. Click through to open the real file, or rewind when the agent takes a wrong turn. ### 💻 Run commands in your terminal Nexrall runs shell commands to install dependencies, run builds, and execute tests — streaming output so you can watch and interrupt. Long-running processes can run in the background, and you can optionally wrap commands in an OS sandbox (Seatbelt on macOS, bubblewrap on Linux). ### ⏪ Checkpoints & rewind Every turn is snapshotted. Use *Nexrall: Rewind Checkpoint* to roll the filesystem and the conversation back to any earlier point, so you can explore risky directions without losing progress. ### 🧩 Sub-agents Nexrall can spawn parallel sub-agents to search the codebase, run tests, or review a diff while it keeps working — each with its own scoped task list, and resumable across turns. ### 🛠️ Custom agents & slash commands Define your own reusable agent types in `.nexrall/agents/*.md` — each with its own tools, model, and instructions — and your own slash commands in `.nexrall/commands/*.md`. They're plain Markdown files that live in your repo, versioned with your code. ### 🔌 MCP tools Extend the agent with any [Model Context Protocol](https://modelcontextprotocol.io) server — local `stdio` commands or remote Streamable HTTP / SSE. MCP connections are shared across chat panels, so opening a second session doesn't spawn duplicate servers. ### 🔐 Permissions that fit you - **Auto mode** — pre-approve every tool and let the agent work unattended. - **Plan mode** — a true read-only lock: the agent can read and plan, but not write or execute. - **Granular rules** — allow or deny specific tools per path. ### 🪝 Hooks Blocking hooks intercept tool calls before they run, so you can gate exactly what the agent is allowed to do in your repo. ### 📜 Audit trail Flip on `nexrall.auditTrail` to record every tool call — and every refused one — to an append-only JSONL file. It's written locally and never leaves your machine, for teams that need a compliance-grade paper trail. ### 🖼️ Attach images & PDFs Attach a screenshot to fix a visual bug, or a PDF to work from a spec. Attachments are gated by model capability, so you're never billed for a turn a model can't see. ### 🧵 Multi-session chat Open independent chat sessions in separate editor tabs from the title-bar icon — each with its own history, checkpoints, and conversation. Sessions can hand work to each other too, with *Nexrall: Message Another Session*. --- ## Getting started 1. Install the extension. 2. Open the Nexrall panel from the Activity Bar. 3. Log in — Command Palette → `Nexrall: Login`. 4. Type a task and go. Try: *"Find why the login flow redirects twice and fix it."* ## Nexrall Code in your terminal The same agent also runs in your terminal — for scripts, CI, and terminal-first workflows: ```bash npm install -g nexrall-code nex "find why the login flow redirects twice and fix it" ``` It shares your Nexrall account and models with the extension. Run `nex --help`, or read more on the [npm page](https://www.npmjs.com/package/nexrall-code). ## Commands | Command | Description | |---|---| | `Nexrall: Open Chat` | Open the chat panel | | `Nexrall: New Chat` | Start an independent chat session in a new editor tab | | `Nexrall: Ask About Selection` | Send selected code to the agent (also in the editor context menu) | | `Nexrall: Explain This File` | Get an explanation of the current file's purpose and structure | | `Nexrall: Rewind Checkpoint` | Roll the workspace and conversation back to an earlier turn | | `Nexrall: Toggle Worktree Isolation` | Run this session in a git worktree so the agent's edits stay out of your working tree | | `Nexrall: Message Another Session` | Send a message to another Nexrall session running on this machine | | `Nexrall: Login` | Sign in to your Nexrall account | | `Nexrall: Logout` | Sign out | ## Settings | Setting | Description | |---|---| | `nexrall.model` | Model to use — Claude, GPT, DeepSeek, or GLM. Defaults to `deepseek-flash`. | | `nexrall.autoApprove` | Pre-approve all tool calls without prompting. | | `nexrall.auditTrail` | Record every tool call (and refused call) to an append-only JSONL audit trail, written locally. | | `nexrall.auditTrailPath` | Where to write the audit trail (default `.nexrall/audit/.jsonl`). | ## Requirements A [Nexrall](https://nexrall.com) account is required — sign up free and manage models and billing from one place, with no API keys to configure. ## Issues & feature requests Bug reports and feature requests are welcome in the public tracker: **https://github.com/Maxrall-Inc/nexrall-issues/issues** Please include your extension version, VS Code version, and OS — it makes reproductions much faster.