#Nexrall

Appearance

Nexrall Code for VS Code ​

Nexrall Code is an AI coding agent embedded directly in Visual Studio Code — built on the same agent engine as the Nexrall CLI and Nexrall Work, exposed as an editor extension.

Install ​

Search "Nexrall Code" in the Extensions view (⇧⌘X on macOS, Ctrl+Shift+X on Windows/Linux), or run:

sh
code --install-extension Maxrall-Inc.nexrall-coding

What it can do ​

  • Read your workspace — semantic navigation (go-to-definition, find-references, hover types) via the language server, not just text search.
  • Multi-file edits — targeted changes across as many files as a task needs, applied atomically.
  • Run commands — a real terminal, including multi-line scripts and heredocs.
  • Live diagnostics — reads the editor's actual error/warning state, not a stale snapshot.
  • Agent loop — plans, edits, runs, observes the result, and continues automatically for as many turns as the task needs.

Working in the editor ​

  • Inline diffs — every edit appears as a diff you can expand right in the chat; click through to open the real file.
  • Checkpoints & rewind — every turn is snapshotted. Nexrall: Rewind Checkpoint rolls the files and the conversation back to any earlier point, so trying a risky direction never costs more than a rewind.
  • Sessions in editor tabs — the Session Manager in the Activity Bar lists your sessions (searchable, with the account & usage summary); Nexrall: New Chat opens an independent session in its own editor tab, with its own history and checkpoints. Nexrall: Message Another Session passes a message to another Nexrall session running on this machine.
  • Sub-agents — the agent spawns parallel sub-agents (searching, running tests, reviewing) while it keeps working; the composer's agents badge tracks the running ones.
  • Worktree isolation — Nexrall: Toggle Worktree Isolation moves this session into its own git worktree (the same mechanism as the CLI's --worktree): its edits never touch your working tree, and writes or shell commands outside it are refused.
  • Commands in a real terminal — install dependencies and run builds and tests with streaming output; long-running processes can move to the background.
  • Attach images & PDFs — drop a screenshot or a spec into a message; attachments are gated by model capability, so you are never billed for a turn a model cannot see.
  • Audit trail — the nexrall.auditTrail setting records every tool call (and every refused one) to an append-only JSONL file, written locally.

Modes ​

  • Auto — the agent edits files directly.
  • Ask — the agent answers questions and explains code without editing anything.
  • Plan — the agent proposes an approach without writing any files; every write tool is hard-blocked in this mode, not just discouraged by instruction.

Models ​

Click the model name in the composer (next to the / button) to open the model picker — models grouped by provider with their per-1M-token prices, plus the thinking-effort slider for the selected model. Nexrall is multi-provider — choose from Anthropic's Claude models (Sonnet 5.5, Opus 5.5, Fable 5.1), OpenAI's GPT-5.6 and GPT-6 families, DeepSeek and GLM.

If you never pick one, the session starts on your account's default — Claude Sonnet 5.5 once your account is unlocked (a first payment, or team access). A never-paid account starts on DeepSeek Flash and sees an Upgrade badge in place of the picker until then.

See Models for the full catalogue and per-1M prices.

Billing: whose wallet pays ​

The Session Manager and the chat panel show your balance, and — if you belong to a team — a Bill to row that switches which budget your usage comes out of. The choice is stored on your account, not in the extension's local config, so it stays in sync with the web app: switching it here switches it everywhere, and vice versa.

A headless or CI run can override it per request, which is how a job opts out of a company budget and bills a personal wallet instead. See Teams & shared budgets.

Safety guardrails ​

Nexrall Code refuses or double-checks a few classes of action automatically:

  • Destructive commands — irreversible operations (force-pushing, dropping a database table, terraform destroy) require explicit confirmation, even if they'd otherwise match an auto-approved pattern.
  • Protected configuration files — writes to files that can run code (shell startup files like .zshenv/.bashrc, git config, package-manager and build-tool configs like .npmrc/.bazelrc, and the agent's own config directories such as .nexrall/) are never auto-approved: they always ask first, even in Auto mode and even under a matching allow rule. See Protected paths.
  • Silent code loss — an edit that would drop existing code behind a placeholder comment like // rest unchanged is refused outright.
  • Test-integrity — weakening a test to make it pass (deleting an assertion, adding .skip) is flagged, not silently accepted as "done."

Updates ​

The extension checks for updates automatically and shows a badge when a new version is available — no manual "check for updates" step needed.

Plugins ​

Nexrall Code supports installable plugins (commands, agents, hooks) — see Plugins.

Commands ​

Every command is available from the Command Palette (⇧⌘P on macOS, Ctrl+Shift+P on Windows/Linux):

CommandWhat it does
Nexrall: Open ChatOpen the chat panel.
Nexrall: New ChatStart an independent session in a new editor tab.
Nexrall: Ask About SelectionSend the selected code to the agent (also in the editor context menu).
Nexrall: Explain This FileExplain the current file's purpose and structure.
Nexrall: Rewind CheckpointRoll the workspace and the conversation back to an earlier turn.
Nexrall: Toggle Worktree IsolationRun this session in a git worktree so its edits stay out of your working tree.
Nexrall: Message Another SessionSend a message to another Nexrall session running on this machine.
Nexrall: Login / Nexrall: LogoutSign in or out.

Reference ​

For the full technical surface, see the reference pages:

Built by Maxrall, Inc.

© 2026 Nexrall. All rights reserved.