Agentic Review

A new agent for reviewing agent-generated code in your editor

Gemini 3 Flash in the Search Subagent

The search subagent is now 3x faster with Gemini 3 Flash

Clickable Diagrams

Mermaid diagrams now link directly to your code

Look At This

Analyze PDFs and images without bloating your context window

Thread Labels

Organize and find your threads with custom labels

Agent Skills

Reuse your existing agent skills in Amp

Amp Python SDK

Amp Python SDK is now available

Find Threads

Find Amp threads by keyword or by which files they touched

Amp, Inc.

Amp is becoming its own company

Opus 4.5: Better, Faster, Often Cheaper

Claude Opus 4.5 is the new main agent model in Amp

Try Claude Opus 4.5 With Us

Test Claude Opus 4.5 in Amp while we evaluate its agentic capabilities

Gemini 3 Pro

Google's Gemini 3 is now the main agent model in Amp

Rush Mode

A faster, cheaper way to use Amp for small, well-defined coding tasks.

Workspace Settings

Configure the Amp CLI per workspace

Social Coding

See how other people are actually using coding agents to do real stuff, with public profiles and threads.

Workspace Entitlements

Per-user usage quotas for Amp Enterprise Premium workspaces

Amp Now Reads Threads

Reference and let Amp read other Amp threads

Command Palette, Not Slash Commands

Amp now has a command palette that replaces slash commands

Review

A new panel for reading agent-generated code

Use Amp Free at Work

The Internet’s best business model (advertising) delivers once again: a $0 coding agent that meets the same stringent infosec standards of Amp’s paid smart mode.

Handoff (No More Compaction)

Move work into new threads without losing your built-up context

More Tools for the Agent

Create new tools for Amp using the CLI instead of writing MCP servers.

Edit, Restore and Fork in the CLI

Edit, restore and fork previous messages in the CLI with Tab

50% Faster Search Agent

Search agent is now 50% faster with optimized parallel execution

The Librarian

A new subagent for searching code on GitHub

Amp Free

Agentic coding is now free for everyone.

Monthly Automatic Top Ups

A new mode for a more predictable Amp budget

Amp TypeScript SDK

GPT-5 Oracle

When Sonnet falls short, GPT-5 can now help as the oracle

Amp CLI connects to VS Code and Neovim

Amp CLI connects to VS Code and Neovim

Now Using Claude Sonnet 4.5

Anthropic released an updated Sonnet 4.5, and Amp now uses it by default

Amp Tab for All

Our free in-editor completion engine for VS Code, Cursor and Windsurf is now on by default for new installs

Globs in AGENTS.md

Streaming JSON

The Amp CLI has two new flags: `--stream-json` and `--stream-json-input` and stuff

Look Ma, No Flicker

Today the Amp CLI reached the next stage in its evolution. It’s no longer just a command-line tool, but a proper terminal user interface.

Bring Your Own Tools

Provide tools to Amp in the form of executable files

1,000,000 Tokens

Amp can now use 1 million tokens of context with Claude Sonnet 4

Custom Slash Commands

Define custom slash commands in the Amp CLI, and use them to insert pre-defined or dynamically-generated prompts

Enterprise Managed Settings

Configure organization-wide settings that override individual settings

MCP Permissions

You can now set `amp.mcpPermissions` to define rules that block or allow MCP servers

JetBrains Diagnostics for Java, Kotlin, and Scala

Read JVM diagnostics from JetBrains IDEs

Multi-File Tab

Amp Tab will now suggest edits across multiple recently-viewed files

Through the Agent, Into the Shell

Execute shell commands, right there in the prompt input

From AGENT.md to AGENTS.md

Amp now uses `AGENTS.md` as the standard filename

A New Model for Amp Tab

Amp Tab uses an updated model with better completion accuracy

New Threads without Leaving the CLI

Start new threads or continue old ones with `/new` and `/continue`

Queue Messages in the CLI

Queue messages with `/queue`

432,000 Tokens

Amp supports 432k-token context with Claude Sonnet 4

Diagnostic-Driven Completions

Amp Tab now uses diagnostics to suggest file-wide fixes

Model Evaluation

Follow our team's real-time evaluation of new frontier models

Try GPT-5 With Us

Test GPT-5 in Amp while we evaluate its agentic capabilities

JetBrains in the CLI

Connect Amp CLI to JetBrains IDEs for context and editing

Tool-Level Permissions

Configure granular permissions for individual tools and arguments

Amp Tab 30% Faster

Amp Tab response times are now 30% faster with optimized infrastructure

Thread Forking in VS Code

We added fork threads to help you explore different directions mid-thread

amp dash x

The Amp CLI has a new flag: `--execute`

Towards a New CLI

Today we released a new version of our Amp CLI

Streamable Transport for MCP

Connect to remote MCP servers with streamable HTTP transport

Multiple AGENT.md Files

Use multiple `AGENT.md` files across directory trees

Oracle

Meet Oracle, Amp's new o3-powered tool for code review and analysis

Better, Faster, Cheaper Summaries

Amp now uses a different model when compacting or summarizing threads that is 6x faster and 30x cheaper

Queued Messages

Queue messages with `Shift+Enter` to send when the agent is idle

Agents for the Agent

Amp now has subagents to handle specialized tasks within threads

Amp Tab

Introducing our new in-editor completion engine that anticipates your next actions

Multi-Root Workspaces

Secret Redaction

Amp now identifies secrets and redacts them so they are not exposed to the LLM, other tools, or ampcode.com

Raising an Agent, Episode 6

Claude 4, sub-agents, and "hot tips" for agentic coding

Terminal Improvements

Commands now inherit shell environments and support interactive mode

TODOs for the Agent

The TODO feature that lets the agent manage its own list of tasks is now enabled by default

Image Support in CLI

Reference images by pasting paths or dragging files into the terminal

Amp Now Uses Claude Sonnet 4

Now that Anthropic has released it, we've switched Amp's primary model to be Claude Sonnet 4

The First 500,000 Messages

Amp Is Now Available To Everyone — Here's How I Use It

As of today, Amp is available to everyone. No more waitlist.

Raising an Agent, Episode 5

Five weeks of Amp evolution and AI coding predictions

History in CLI

Navigate message history with arrow keys and keyboard shortcuts

No More BYOK

The best product is built by iterating fast at the model↔product frontier, so we removed Isolated Mode

AGENT.md

Configure Amp with project guidance in `AGENTS.md` files

File Mentions in CLI

Fuzzy-search files with `@` mentions in interactive mode

Manual

Read the comprehensive operator's guide to using Amp effectively

How to Build an Agent

Learn how to build a fully functioning code-editing agent

Frequently Ignored Feedback

Our responses to common feedback we are intentionally not acting on

Raising an Agent, Episode 4

What will AI do to open-source? What does it mean for GitHub? What does it mean for interviewing engineers?

Raising an Agent, Episode 3

What does this all mean for code search? How do you balance coding knowledge vs. letting the AI do it?

Raising an Agent, Episode 2

Is the magic with these agents that there are no token limits? Where do the agents fail? Do they need guidance?

Raising an Agent, Episode 1

Build your Amp agent from the ground up

200k Tokens Is Plenty

Short threads are cheaper, better, and easier to reason about.

How We Think about Permissions

Agents work best when they can get feedback about the changes they are making.

Agents for the Agent

Amp now has new subagents and they're more powerful than ever.

Amp is now available. Here's how I use it.

How I personally use Amp to develop software.

How We Build

How we build Amp.

Frequently Ignored Feedback (FIF)

Our responses to common feedback that we are intentionally not acting on.