Short, direct answers to the questions people ask most. Each answer links to the page that covers it in full. For step-by-step fixes to specific problems, see Troubleshooting.

Getting started

100xprompt is a premium AI coding agent that lives in your terminal, browser, and IDE. It reads your codebase, writes and edits code, runs commands, and connects to your other tools - driven by frontier models. You describe the outcome; it does the engineering work.
The fastest path is the one-line installer:
curl -fsSL https://100xprompt.com/install | bash
You can also install through npm, bun, pnpm, yarn, Homebrew, or Scoop. Full instructions, system requirements, and upgrade/uninstall steps are on the Installation page.
No. Move into any project and run 100xprompt. On first launch you’re guided to sign in to 100xprompt, and you can start working right away. See the Quickstart.
Launch 100xprompt in your project directory and describe what you want in plain language - “add pagination to the users endpoint,” “find where rate limiting is handled,” or “fix the failing tests.” It reads the relevant files, proposes a plan, and gets to work. See How It Works.
100xprompt runs on two in-house frontier models: Flash for fast, lightweight work and Pro for deep reasoning and large refactors. Switch between them any time, or run 100xprompt models to see the exact catalog you can select right now. See Models.
Yes. Open the Switch model picker inside a session and choose a new model. It takes effect immediately - no restart, no lost context. You can also set a default model in 100xprompt.json, or pass --model 100xprompt/pro to a single run.
Run 100xprompt auth login, choose 100xprompt, and paste the API key from your account at 100xprompt.com. Your credential is stored securely and reused automatically for every session. See Authentication.
Point lightweight background steps (like session titles and short summaries) at Flash with the small_model key, while your main work stays on Pro. You can also assign a model to individual agents - see per-agent model choice on Models.
It reads your codebase, edits and writes files, runs shell commands and tests, drives multi-step tasks, delegates work to parallel subagents, and connects to external systems through MCP. See the Capabilities overview.
Yes. It runs shell commands, tests, and build steps as part of completing a task. Every command and file edit is governed by permissions, so you stay in control of what runs. See Permissions.
Every task is handled by an agent - a focused operator with its own instructions, model, and allowed tools. Any agent can spin up subagents to fan work out in parallel, each working in its own isolated context and reporting back a result. See Agents & Subagents.
Project rules live in AGENTS.md, and 100xprompt maintains its own long-term memory across sessions so preferences and hard-won context carry forward. See Memory & Context and Rules.
Your code stays on your machine. 100xprompt runs locally and sends the model only the context it needs to answer your request. That model traffic goes to 100xprompt’s hosted service over an encrypted connection. See Security.
Actions that change your system - editing files, running commands, invoking tools - are gated by a permission rule that resolves to allow, ask, or deny. By default you’re asked before anything consequential runs, and you can approve once or always. Permission modes let you loosen or tighten this per session. See Permissions.
Yes. Define permission rules in 100xprompt.json to allow or deny specific tools and patterns, and switch to plan mode when you want read-only investigation with no changes at all. See Permissions.
Provider credentials are stored securely on your machine. Review connected accounts with 100xprompt auth list and remove one with 100xprompt auth logout.
Two main paths. To connect external systems - issue trackers, docs, databases, browsers - register an MCP server; its capabilities appear as regular tools automatically. To package a repeatable workflow, write a Skill that bundles instructions and helper files. See Connect Tools & Data with MCP and Skills.
A Skill teaches 100xprompt a repeatable procedure it can invoke when relevant. A Command is a reusable prompt you trigger with a slash (like /init). A Plugin bundles skills, commands, agents, and MCP servers into one installable package. See Skills, Commands, and Plugins.
Yes. Publish plugins to a Marketplace so your team installs them with one command. See Marketplace.
Yes. Use 100xprompt run "your task" for a single headless run - ideal for pipelines. Add --format json for machine-readable output, --model 100xprompt/pro to pin a model, --agent <name> to choose an agent, or --continue / --session <id> to resume prior work.
100xprompt run --format json --model 100xprompt/pro "Update the changelog for the latest release"
100xprompt integrates with Visual Studio Code (and VS Code Insiders), Cursor, and Windsurf. See IDE.
Yes. Run 100xprompt web to start the server and open the browser interface - the same agent in a full graphical view. See Web.
Share a session to get a link others can open in their browser. Trigger the share action inside a session, or pass --share to 100xprompt run. To set the default, set share to manual, auto, or disabled in 100xprompt.json. See Sharing.
Yes. The SDK lets you drive 100xprompt from your own applications and scripts. See SDK.

Glossary

Clear definitions for every 100xprompt term used across these docs.

Troubleshooting

Fixes for install, sign-in, model, permission, and connection issues.

Best Practices

Habits that get the most out of 100xprompt on real projects.

Quickstart

Run your first task and see 100xprompt work on real code.