# Introduction

Hoplite is a coding agent platform. You connect a GitHub repository, describe a task, and an agent does the work in a real, isolated dev environment — editing code, running your tests, starting your app, checking its work in a browser, and opening a pull request when it's done.

<Callout type="tip" title="TL;DR">
Sign in at [hoplite.sh](https://hoplite.sh), connect GitHub, pick a repository, and send your first message. The [Quickstart](/docs/quickstart) takes you from there to a merged pull request.
</Callout>

## The mental model

Four terms carry the whole product: **project → thread → sandbox → pull request**.

| Term | What it is |
| --- | --- |
| **Project** | A GitHub repository binding plus everything scoped to it: base branch, environment variables, scripts, model defaults, and instructions. |
| **Thread** | One conversation and one task against a project. Each of your messages kicks off a **run** — a single agent execution turn. |
| **Sandbox** | The isolated dev environment created for each thread, with your repo cloned, your setup script applied, and your environment variables injected. |
| **Pull request** | The output. The agent opens a PR from the thread's diff, and the PR's checks, reviews, and comments feed back into the conversation. |

Projects live in a **workspace** — your team's account, which holds members, billing, and integrations. You watch the agent's work stream into the thread, approve sensitive actions like file edits and shell commands, and steer with follow-up messages at any time.

## What do you want to do?

### Start

- **[Quickstart](/docs/quickstart)** — go from a fresh account to a merged, agent-authored pull request.
- **[Make your repo agent-ready](/docs/agent-ready)** — add setup and run scripts, environment variables, and instructions so agents work well in your codebase.
- **[Import from another tool](/docs/import)** — bring your Claude Code, Codex, or OpenCode history and skills with you.
- **[Hand off a live session](/docs/cli/handoff)** — continue the local agent session you're in right now as a cloud thread.
- **[Desktop app](/docs/desktop)** — the native macOS client, with an optional local host for Codex and Claude Code.

### Ship

- **[What to delegate](/docs/delegate)** — which tasks suit an agent thread, how to size them, and how to run threads in parallel.
- **[Write effective instructions](/docs/prompting)** — state the goal, the constraints, and how the agent should verify its work.
- **[Prompt templates](/docs/prompt-templates)** — copy-paste starting points for bugs, features, refactors, and more.
- **[Threads](/docs/threads)** — the conversation UI: approvals, queued messages, previews, and the PR rail.
- **[GitHub & pull requests](/docs/github)** — draft PRs, the review loop, and merging without leaving the thread.

### Automate

- **[Automations](/docs/automations)** — run agents from webhooks and schedules.
- **[Slack](/docs/integrations/slack)** and **[Linear](/docs/integrations/linear)** — connect Hoplite to the tools where your team already works.
- **[Agent skills](/docs/agent/skills)** — reusable instruction bundles the agent loads on demand.
- **[API](/docs/api)** — drive Hoplite from scripts with workspace API keys.

Introduction

Hoplite is a coding agent platform: delegate work to agents that run in real dev environments and ship pull requests.

Hoplite is a coding agent platform. You connect a GitHub repository, describe a task, and an agent does the work in a real, isolated dev environment — editing code, running your tests, starting your app, checking its work in a browser, and opening a pull request when it's done.

Sign in at hoplite.sh, connect GitHub, pick a repository, and send your first message. The Quickstart takes you from there to a merged pull request.

The mental model

Four terms carry the whole product: project → thread → sandbox → pull request.

Term	What it is
Project	A GitHub repository binding plus everything scoped to it: base branch, environment variables, scripts, model defaults, and instructions.
Thread	One conversation and one task against a project. Each of your messages kicks off a run — a single agent execution turn.
Sandbox	The isolated dev environment created for each thread, with your repo cloned, your setup script applied, and your environment variables injected.
Pull request	The output. The agent opens a PR from the thread's diff, and the PR's checks, reviews, and comments feed back into the conversation.

Projects live in a workspace — your team's account, which holds members, billing, and integrations. You watch the agent's work stream into the thread, approve sensitive actions like file edits and shell commands, and steer with follow-up messages at any time.

What do you want to do?

Start

Quickstart — go from a fresh account to a merged, agent-authored pull request.
Make your repo agent-ready — add setup and run scripts, environment variables, and instructions so agents work well in your codebase.
Import from another tool — bring your Claude Code, Codex, or OpenCode history and skills with you.
Hand off a live session — continue the local agent session you're in right now as a cloud thread.
Desktop app — the native macOS client, with an optional local host for Codex and Claude Code.

Ship

What to delegate — which tasks suit an agent thread, how to size them, and how to run threads in parallel.
Write effective instructions — state the goal, the constraints, and how the agent should verify its work.
Prompt templates — copy-paste starting points for bugs, features, refactors, and more.
Threads — the conversation UI: approvals, queued messages, previews, and the PR rail.
GitHub & pull requests — draft PRs, the review loop, and merging without leaving the thread.

Automate

Automations — run agents from webhooks and schedules.
Slack and Linear — connect Hoplite to the tools where your team already works.
Agent skills — reusable instruction bundles the agent loads on demand.
API — drive Hoplite from scripts with workspace API keys.