New · v1.8.72 · Full Access mode · See what's new →

The Code Agent that perceives what you're doing

Runs 100% on your machine. Perceives · Anchors · Calibrates. Your code never leaves.

Download for Windows
macOS · other platforms
Free · No credit card · Runs locally · By downloading you agree to RSANC-1.0 NonCommercial
app.tsxutils.ts+
1import { useState } from "react";
2import { apiClient } from "./api";
3
4export function Dashboard() {
5 const [rows, setRows] = useState<Row[]>([]);
6 const [loading, setLoading] = useState(false);
7
8 async function refresh() {
9 setLoading(true);
10 const next = await apiClient.list();
11 setRows(next);
REFACTOR DASHBOARD
Extract each row in Dashboard.tsx into a standalone DashboardRow component. Preserve selection state, keyboard navigation, and virtualization behavior.
readDashboard.tsx · useSelection.ts
createdDashboardRow.tsx +62
editedDashboard.tsx +18 −74
▸ Mapped the Dashboard layout ✓
Sessions
Refactor Dashboard componentthinking
Wire crash reporter to backend+42
Migrate MemoryDashboard tabs+112 −58
3
cognitive principles
227 MB
Windows installer
5 / 8s
static / rendered fetch
0
code leaks to cloud
USE CASES

Coding is just the opening act.

One agent. One desktop. Holds every tool you'd otherwise switch to during the day.

01

Ship code

Build Feature · Debug Production · Refactor · UI Design '4 workbenches out of the box.

02

Do research

Research panel + preset literature and search templates. Prove an idea into a plan.

03

Learn anything

9 tutor packs 'math, language, coding, science, humanities, exam prep, arts. Each with level and pedagogy.

04

Track markets

Twin brain dual view 'entity graph + map vs forecast + sentiment + impact. Bitcoin and Red Sea shipping live.

05

Own your box

Ops panel five up: Health · Services · Disks · Logs · Startup. Reads sysinfo live.

06

Prototype live

Push interactive demos to the Stage canvas. 3D · dashboard · particles · chat UI.

AUTONOMOUS

Agents that build for you

Hand off tasks to a code agent that autonomously plans, reads your workspace, calls tools, edits files, and runs commands to iterate until the job is done.

Every write is diff-approved. Every command is confirmed. You stay in control at the frame boundaries 'the agent stays in the loop between them.

Get started with agents →
batch.go
1package http
2
3import (
4 "context"
5 "encoding/json"
6 "errors"
7 "io"
8 "log/slog"
9 "net/http"
10)
11
12type Result[T any] struct {
13 Name string `json:"name"`
14 Value T `json:"value,omitempty"`
BATCH IMAGE PROCESSING
Add a POST /process/batch endpoint. Process up to 8 files in parallel with per-request budget, structured logging, and OpenTelemetry spans.
readserver.go · processor.go
createdbatch.go +128
editedserver.go +4 −0
createdbatch_test.go +96
▸Mapped the service layout
📂

Filesystem

list_dir · read_file · write_file · glob · grep 'every write diff-approved.

⚡

Shell

run_command with pre-flight guard · start_background_task for long-running.

🧠

Memory

22 recall tools · SAG streaming · never sends chat off-device.

🌐

Web

web_search · browser_fetch via Rust ureq 'CN network friendly.

📋

Plan

agent_plan_create · step-by-step visible progress in a side panel.

🎨

Canvas

Interactive HTML preview · math · geometry · chemistry · games.

🔌

Plugins

Plugin marketplace · minisign verified · verifiable tool packages.

🎯

Approval

Per-category policy · ask / auto / deny · session-scope grants.

📊

Usage

Per-provider cost tracking · monthly budget alert · full history.

80+ tools ready to compose · See the full catalog →
COMPOSABLE

Tools built for real work

Extend your agent with tools from the plugin marketplace, or write your own with the plugin SDK to power your team's unique scenarios.

Every tool ships with an OpenAI-format schema, a JSON Schema for validation, and an approval flow that fits your policy.

Browse the plugin catalog →
UNIVERSAL

Code in any language

Provider-agnostic. Whether you pay for OpenAI, Anthropic, DeepSeek, Zhipu, Moonshot, Alibaba DashScope, or run Ollama locally 'the agent adapts to what you configure.

Multi-model: architect / editor / reviewer can each be a different model. Cheap for planning, powerful for editing, cautious for review.

Configure providers →
JSJavaScript
TSTypeScript
PyPython
GoGo
RsRust
JaJava
C+C++
C#C#
RbRuby
KtKotlin
SwSwift
HTHTML
SQSQL
MDMarkdown
YMYAML
PRINCIPLES

Four commitments your code depends on

These aren't features. They're constraints we optimize the whole product around, in every release.

01

Local-first

Chat routes direct to your provider. Your code, your history, your memory never touch our servers.

02

Perception-first

The agent reads before it acts. list_dir, read_file, memory_recall 'evidence over guesses.

03

Approval-gated

Every write, every command, every risky action. You confirm the frame; agent runs the loop.

04

Signed updates

Every release is minisign-signed. Your app verifies before it installs. No silent tampering.

CAPABILITIES

Put the agent inside your own machine

Full Access · 3 tiers
Workspace · mounted · full-disk. System dirs stay hard-denied.
Local-first
Code and traces never leave your machine. Only LLM calls go out.
FetchOrchestrator
4-lane fetch · dual verifier · auto-promoted after 5 jobs at 80%+.
World-model cache
Cache hits skip web and LLM. 35'5 seconds saved per run.
SCREENSHOTS

This is what opens.

What you install is what you see.

Education · themed learning packs
Education · themed learning packs
Preset tutors for math / language / coding / science / humanities / exam prep / arts 'each with persona, level, and format tuned.
Task Center · long-running work
Task Center · long-running work
Active / paused / completed / failed at a glance. Deep and Auto-Work runs are fully re-playable step-by-step.
My Skills · 18 preset skills
My Skills · 18 preset skills
Build Feature, Debug Production, Refactor, UI Design and more 'every skill declares its expected turns and cost budget.
Intelligence · local Ops surface
Intelligence · local Ops surface
Health / Services / Disks / Logs / Startup piped straight from sysinfo 'real CPU, real processes, real disks, no mocks.
Intelligence · Twin brain 2×2
Intelligence · Twin brain 2×2
Same topic, two lenses: Scene (entity graph, geospatial) and Boards (forecast, sentiment, impact). Bitcoin and Red Sea shipping cases render live.

Download free. Run it now.

Free · no credit card · runs locally · your code stays on your machine.

Download for Windows
macOS