# An IDE that remembers your codebase — Hanzo AI

> AI-first IDE powered by a high-performance Rust engine. Persistent project memory, call-graph awareness, and sandboxed tool execution.

Hanzo IDE · Developers

# An IDE that remembers your codebase

The editor you already know, on a Rust engine that is ours. The agent keeps a memory of your project across sessions, reads the call graph before it changes a line, and runs every tool in a sandbox.

[Open the IDE](https://ide.hanzo.ai)[View on GitHub](https://github.com/hanzoai/ide)

Open source (Apache 2.0). The VS Code workbench on a Rust engine. Open VSX extensions run as they are.

IDE — Hanzo

[CLI](https://hanzo.ai/cli)[Playground](https://hanzo.ai/playground)[Desktop](https://hanzo.ai/desktop)[API Keys](https://hanzo.ai/cloud/api-keys)

- $hanzo
- >deploy the gateway to staging
- ·reading the project
- ✓gateway is on staging

`api.hanzo.ai/v1/code`[Open in the console](https://hanzo.ai/home)

Example project. Every name in it is one you would give your own.

Paste and ship

Paste this into any agent. It reads the skill manifest and calls IDE from there.

Agent promptapi /v1/codeCopy

```
Read https://hanzo.ai/skill.md and use Hanzo IDE in my project. Start with: hanzo code
```

What is Hanzo IDE

## The workbench you know, on an engine that is ours

The editor surface, the extension model and the keybindings are the ones you use today. Everything under them is Rust: the agent loop, the memory, the indexing, the sandbox.

01

### It remembers

Engram gives the agent a memory that lasts past the conversation: what happened, what is true about the project, and how things are done here. Memories are consolidated, decay when unused, and come back when they matter.

02

### It understands structure

Tree-sitter indexing gives the agent the call graph, the type hierarchy and the impact radius of a change, so it reads the code that matters instead of every file.

03

### It is safe by default

Tool calls run in a sandbox. Secrets live in the keychain, sensitive fields are encrypted at rest, and every action lands in an audit trail. All of it is in the core, none of it is a plugin.

Capabilities

## What the window carries

### Three tiers of memory

A sensory buffer for the turn, working memory for the session, and a long-term graph of episodic, semantic and procedural memory that survives a restart.

### Recall that ranks

Full-text and vector search run together across every store, fused and reranked, and every answer arrives with a quality score.

### Agent profiles

Name what an agent may touch, which model it runs on, and how far it may go on its own. Switch profiles per task.

### An activity feed

Every tool call, every decision and every approval the agent made, in real time, in one column.

### Open VSX extensions

The extensions you rely on install from Open VSX and run natively. Language servers, themes, keymaps.

### Ten layers of security

PII detection, field-level encryption, sandbox enforcement, keychain integration and audit trails, built into the engine.

Get started

## Open it in the browser, or build it on your machine

bash

```
# Open it in the browser open https://ide.hanzo.ai # Or build the desktop app from source git clone https://github.com/hanzoai/ide.git cd ide npm install npm run tauri:dev # First launch: pick a provider, paste a key, choose a model.
```

[Open the IDE](https://ide.hanzo.ai)[View on GitHub](https://github.com/hanzoai/ide)

## Open the IDE. It already knows the way around.

Your project, its call graph, and what the agent learned last week, in one window.

[Open the IDE](https://ide.hanzo.ai)[GitHub](https://github.com/hanzoai/ide)[Hanzo Cloud](https://platform.hanzo.ai)

Related capabilities

[CLI](https://hanzo.ai/cli)[Playground](https://hanzo.ai/playground)[Desktop](https://hanzo.ai/desktop)

## Get IDE

[Deploy to Cloud](https://platform.hanzo.ai/deploy)