Autohand Code

# Ship deployable changes faster.

Autohand plans, edits, tests, and reviews your code across the tools your team already uses.

[Start Pro](https://autohand.ai/signup/?return=https%3A%2F%2Fconsole.autohand.ai%2Fupgrade%3Ftier%3Dpro%26billing%3Dmonthly%26from%3Dcode)[See plans](https://autohand.ai/code/pricing/)

![](https://autohand.ai/images/hero-media/59ef2ebb6f8b-1200.webp)

## Choose where you want to work

Autohand Code works wherever you code. Pick your preferred environment.

### CLI

Terminal-first workflow for power users

### VS Code

Full IDE integration with inline suggestions

![Zed Editor](https://autohand.ai/logos/zed_editor/256px-Zed_Editor_Logo.png)

### Zed Editor

Native integration for Zed users

![JetBrains](https://autohand.ai/logos/jetbrains/jetbrains-icon.svg)

### JetBrains

IntelliJ, WebStorm, PyCharm and more

### Chrome

Browser extension for code reviews

### Slack

AI assistance in your workspace

![Autohand Squad](https://autohand.ai/images/squad/app-icon.png)

### Squad

Autohand Squad App for agent teams

Agent SDK

## One SDK.  
Every language.

Build review bots, CI agents, and internal dev tools on the same agent loop that powers Autohand Code — in the language your team already ships.

[Read SDK docs](https://docs.autohand.ai/agent-sdk/)[Quickstart](https://docs.autohand.ai/agent-sdk/quickstart/)

9+

SDK languages

Streaming

Event-based API

Self-host

Or use the cloud

agent.ts

```
import { Agent, AutohandSDK } from '@autohandai/agent-sdk'

const sdk = new AutohandSDK({ cwd: repo })
await sdk.start()

const agent = Agent.fromSDK(sdk)
const result = await agent.run('Review this branch and patch risks')

render(result.text)
```

## Recommended for coding agents

High-performance models optimized for coding tasks with tool support, prompt caching, and Autohand agent orchestration.

[**Autohand Fantail**_AUTOHAND_`fantail`Context: **256K**Fast coding loops, completions, reviews, and short agent tasks.](https://autohand.ai/models/fantail/)[**Autohand Moa**_AUTOHAND_`moa`Context: **1M**Deep reasoning for larger refactors, planning, and multi-step changes.](https://autohand.ai/models/moa/)[**GLM-5.1**_SOTA_`glm-5.1`Context: **200K**Best for long-horizon agentic engineering.](https://autohand.ai/docs/integrations/zai.html)[**GLM-5.2**_1M_`glm-5.2`Context: **1M**Best for 1M-context long-horizon agentic engineering.](https://autohand.ai/docs/integrations/zai.html)[**Sakana AI Fugu**_FAST_`sakana-ai-fugu`Context: **1M**Balanced low-latency multi-agent orchestration for coding and chat.](https://autohand.ai/docs/integrations/sakana.html)[**Sakana AI Fugu Ultra**_ORCHESTRATOR_`sakana-ai-fugu-ultra`Context: **1M**Best for multi-agent orchestration across frontier models.](https://autohand.ai/docs/integrations/sakana.html)[**Gemma 4 12B**_12B_`gemma-4-12b`Context: **256K**Best laptop-ready Gemma for local coding agents.](https://autohand.ai/docs/models/)[**Gemma 4 26B A4B**_MOE_`gemma-4-26b-a4b`Context: **256K**Best Gemma MoE for fast local agentic workflows.](https://autohand.ai/docs/models/)[**Kimi K2.6**_CHEAP_`kimi-k2.6`Context: **256K**Best for agent swarms and long autonomous runs.](https://autohand.ai/docs/models/)[**DeepSeek V4 Pro**_1M_`deepseek-v4-pro`Context: **1M**Best for 1M-context and competitive coding.](https://autohand.ai/docs/integrations/deepseek.html)[**Qwen3-Coder-Next**_EFFICIENT_`qwen3-coder-next`Context: **256K**Best efficiency per active parameter.](https://autohand.ai/docs/models/)[**Qwen3.8-Max**_1M_`qwen3.8-max`Context: **1M**Qwen flagship for long-horizon agentic coding.](https://autohand.ai/docs/models/)[**Qwen3.8-Max-0902**_PINNED_`qwen3.8-max-0902`Context: **1M**Dated snapshot of Qwen3.8-Max for repeatable runs.](https://autohand.ai/docs/models/)[**Qwen3.8-27B**_DENSE_`qwen3.8-27b`Context: **1M**Best dense model for repo-level coding.](https://autohand.ai/docs/models/)[**MiniMax M2.5**_FREE_`minimax-m2.5`Context: **200K**Best free hosted open-weight coding model.](https://autohand.ai/docs/models/)[**GLM-5**_SELF-HOST_`glm-5`Context: **200K**Best foundation for local self-hosting.](https://autohand.ai/docs/integrations/zai.html)[**Devstral 2**_MISTRAL_`devstral-2`Context: **256K**Best Mistral coding model.](https://autohand.ai/docs/models/)[**Devstral Small 2**_GPU_`devstral-small-2`Context: **128K**Best for consumer GPUs.](https://autohand.ai/docs/models/)[**Trinity Large Thinking**_REASONING_`trinity-large-thinking`Context: **128K**Best US-origin open reasoning model.](https://autohand.ai/docs/models/)

[View all coding models](https://autohand.ai/models/)

Capabilities

## Built for autonomous coding at scale

Autohand Code goes beyond autocomplete. It runs agent teams, composes with Unix pipes, and evolves your codebase through structured planning.

Project Analysis

📁Indexed 1,247 files

📦Found 45 dependencies

🔗Mapped 892 imports

✨Detected: TypeScript + React

01

### Deep codebase understanding

Autohand Code indexes your entire project to understand architecture, dependencies, and coding patterns. Every suggestion fits your existing codebase.

-   Reads all files in your project
-   Understands imports and dependencies
-   Follows your coding conventions

02

### Agent Teams

Orchestrate multiple agents working in parallel on complex tasks. A lead agent delegates work to specialized teammates - code writers, testers, researchers, and reviewers - with automatic crash recovery and task reassignment.

-   Lead-teammate architecture with task delegation
-   6 built-in agent types for specialized work
-   Crash recovery with automatic task reassignment

Agent Teams

leadTeam Lead: Planning auth refactor

devCode Writer: Implementing OAuth2

testTester: Running integration tests

revReviewer: Pending code review

Auto Mode

✓Reading codebase (worktree isolated)

✓Planning changes

⋯Implementing feature (iteration 3/10)

○Running tests + circuit breaker check

○Creating PR

03

### Autonomous task execution

Describe a complex task and let Autohand handle it end to end. Auto-mode runs in a git worktree for safety, with a circuit breaker that stops unproductive loops, cost limits, and checkpoint intervals.

-   Git worktree isolation protects your main branch
-   Circuit breaker stops loops and wasted compute
-   Cost limits, runtime caps, and checkpoint intervals

04

### Unix pipe composability

Autohand fits into your existing Unix workflows. Pipe data in from any command, get structured output back. Use it with xargs, jq, and standard shell utilities for scripting and automation.

-   Smart stdin detection for piped input
-   JSON output with --json for programmatic use
-   Works with xargs, jq, and shell scripts

Pipe Mode

$git diff | autohand 'explain changes'

$cat error.log | autohand 'find root cause'

$autohand --json 'list todos' | jq '.items\[\]'

→Structured JSON output for automation

Plan Mode

\[PLAN\]Searching codebase for auth patterns

\[PLAN\]Reading existing middleware

✎Presenting approach for approval

\[EXEC\]Implementing approved plan

05

### Plan before you execute

Toggle Plan Mode with Shift+Tab to review the agent's approach before any code changes. The agent uses read-only tools during planning, then gets full access after you approve the strategy.

-   Read-only planning phase for safe exploration
-   Approve, reject, or refine before execution
-   Three approval modes: manual, auto-accept, or clear context

06

### Modular skills system

Skills are reusable workflow packages that teach the agent how to handle specific tasks. Use built-in skills, install from the community, or auto-generate custom skills tailored to your project with a single flag.

-   Auto-skill generation from project analysis
-   Community skill registry with drag-and-drop install
-   Composable with agent teams for structured workflows

Skills

★frontend-ui: React component patterns

★tdd: Test-driven development workflow

+auto-generating 3 project-specific skills

★sre-ops: Incident response runbook

MCP Servers

●github: Connected (stdio)

●postgres: Connected (SSE)

●brave-search: Connected (HTTP)

+/mcp install sequential-thinking

07

### Extensible with MCP

Connect external tools, databases, and APIs through the Model Context Protocol. Install from a curated registry of 12+ servers or add your own. Servers connect asynchronously in the background so you never wait.

-   Curated registry: GitHub, Postgres, Brave Search, Slack, and more
-   Non-blocking startup for instant responsiveness
-   Namespaced tools prevent collisions between servers

08

### Privacy-first design

Connect to local models for complete data sovereignty, or use our secure cloud with zero data retention. Your code is never stored or trained on.

-   Local model support (Ollama, llama.cpp, MLX)
-   Zero data retention policy
-   SOC 2 Type II certified

Privacy Settings

Use local models (Ollama, MLX)

Zero data retention

Encrypt in transit

✓No code stored

Git Integration

⎇Created branch: feature/auth-flow

+Staged 4 files

●Commit: Add OAuth2 authentication

↑Pushed to origin

⊞Worktree: isolated experiment branch

⊕PR #142 created

09

### Built-in git workflows

Autohand handles your git operations seamlessly. Create branches, stage changes, write commits, open pull requests, and isolate risky work in git worktrees automatically.

-   Smart commit message generation
-   Git worktree isolation for safe experimentation
-   One-command PR creation

40+Slash commands

50+Built-in tools

50+Programming languages supported

15Locale languages

English, Spanish, French, German, Italian, Portuguese, Dutch, Russian, Chinese (Simplified), Chinese (Traditional), Japanese, Korean, Arabic, Hindi, Polish

What developers say

## Trusted by developers worldwide

"Autohand Code cut our refactoring time by 80%. It understands our codebase better than most junior devs."

WT

Wei Lin TanTech Lead at Grab

"The auto mode is incredible. I described a feature and it built the whole thing, including tests."

YT

Yuki TanakaSenior Engineer at Mercari

"Finally an AI coding tool that respects privacy. Local model support was a deal-breaker for us."

PS

Priya SharmaCTO at Razorpay

"I shipped a complete authentication system in 2 hours. This tool is a game-changer for solo developers."

JP

Jihoon ParkIndie Developer, Seoul

"The CLI workflow fits perfectly into my terminal-centric setup. No context switching needed."

AW

Arief WibowoDevOps Engineer at Gojek

"Our team onboarding dropped from 2 weeks to 3 days. Autohand helps new devs understand our codebase fast."

MK

Min-Ji KimEngineering Manager at Coupang

"The multi-file refactoring saved us from a painful manual migration. Worth every penny."

RP

Raj PatelPrincipal Engineer at Flipkart

"I use it with Ollama for completely offline coding. Perfect for sensitive projects."

MC

Mei ChenSecurity Engineer at DBS Bank

"Autohand understands our monorepo structure instantly. Saved us weeks of context-building with new hires."

SM

Sophie MitchellStaff Engineer at Notion

"The code quality is production-ready. It follows our design system patterns without being told."

JC

James CooperFrontend Lead at Schneider Electric

"We integrated Autohand into our CI pipeline. It catches issues before they hit code review."

EN

Emma NgataPlatform Engineer at Contact Energy

"Building rocket software requires precision. Autohand delivers clean, testable code every time."

TW

Tane WilliamsSoftware Engineer at Rocket Lab

## Built for |

Work with Autohand **code cli** directly in your codebase. Build, debug, and ship from your terminal, IDE, or Slack. Describe what you need, and Autohand handles the rest.

`curl -fsSL https://autohand.ai/install.sh | sh`

`bun add -g autohand-cli`

Autohand Code pricing

## Start free. Add capacity when the work grows.

Compare hosted credits, model access, agent workflows, team controls, and enterprise deployment options.

[Compare every plan](https://autohand.ai/code/pricing/)

## Frequently asked questions

Autohand Code supports VS Code, JetBrains IDEs (IntelliJ, WebStorm, PyCharm, and more), and Zed Editor with full IDE integration. JetBrains is available via the ACP protocol registry. You can also use the CLI for terminal-based workflows, Unix pipe scripting, and CI/CD automation.

Agent Teams let you orchestrate multiple agents working in parallel on complex tasks. A lead agent breaks down the work and delegates to specialized teammates like code writers, testers, researchers, and reviewers. Tasks have dependencies and crashed agents get their work automatically reassigned.

Pipe Mode lets you use Autohand in Unix pipelines. Pipe data in from any command (like git diff or cat error.log), and get structured output back. Use the --json flag for programmatic JSON output that works with tools like jq and xargs.

Plan Mode separates thinking from doing. Toggle it with Shift+Tab and the agent switches to read-only tools for exploration and planning. Once you approve the approach, it gets full access to implement the plan. You can auto-accept, manually approve each step, or clear context between phases.

Yes. You can connect Autohand Code to local LLMs like Ollama, llama.cpp, or MLX (Apple Silicon) for a fully offline and private coding experience.

By default, we use secure cloud providers for the best performance. However, your code is never stored or trained on. You can also switch to local-only mode for complete data sovereignty.

Autohand Code is a self-evolving code agent that spans CLI, IDE, and Slack in a single platform. It offers agent teams for parallel work, pipe mode for Unix composability, a skills system for reusable workflows, plan mode for safe exploration, and MCP support for connecting external tools. No other tool combines all of these in one agent.

Source: https://autohand.ai/code/