Core Concepts

A quick map of the vocabulary used across Nexrall's docs.

Agent

An agent is a loop that reasons, calls tools (read/write files, run commands, call APIs), observes the result, and repeats until the task is done or it needs your input. Nexrall runs several agents, each shaped for its surface:

• Fullstack Agent Coding — the web agent that reads/writes an app's file tree, runs commands, and tests its own work before deploying.
• Nexrall Code, the Nexrall CLI, and Nexrall Work — the developer tools, which share one agent engine for reading and editing files on your own machine.
• The Canvas agent — writes a single self-contained document/artifact (report, deck, chart) rather than a full app.
• A sub-agent — a narrower, specialised agent the main agent delegates to, scoped to only the tools and permissions it needs (review, testing, research, migration).

Session

A session is one continuous conversation with an agent — the chat history, any files or artifacts it produced, and its progress ledger (see below). Sessions persist so you can close the tab and come back later without losing context.

Progress ledger

Long-running agent work is prone to "context rot" — summarizing a summary until the original goal gets fuzzy. Nexrall's agents maintain a progress ledger: a structured, machine-tracked record of what's been touched (files edited, tests run and their pass/fail result) that survives context compaction, pinned next to the original task description. This is what lets an agent work through dozens of turns without drifting or "forgetting" a failed test.

Tools & connectors

Agents act through tools — file edits, shell commands, HTTP requests — and connectors, which are pre-built integrations (Slack, GitHub, Google Sheets, Discord, and more) an agent can call without you wiring authentication by hand.

Sub-agents & plugins

A sub-agent is a specialist agent the main agent delegates to for a focused piece of work — each runs in its own context with its own tool allowlist and model. Plugins bundle Skills, MCP connectors and sub-agents together as one installable unit, so capabilities can be shared and versioned as a single package.

Models

Nexrall is multi-provider — every agent can run on models from several providers: Anthropic (Claude Sonnet 5, Opus 5, Fable 5), OpenAI (GPT-5.4, GPT-5.4 Mini, GPT-4.1), DeepSeek (V4 Pro, V4 Flash), and Qwen (Qwen3.7 Max). Pick the model that fits the task and your budget from the picker in any surface.

See Models for the full catalogue, context windows, relative cost, and how to switch.

Nexrall Work

Nexrall Work is the desktop app for the coding agent — a terminal-free, chat-first window that works across the folders you connect and can create and edit Word, Excel and PowerPoint files. It runs the same agent as Nexrall Code and the CLI. See Nexrall Work.

Wallet

Every account has one wallet balance, spent on AI usage (coding, Canvas, and image generation). See Billing & Wallet.