Adze

One agent for your whole product team

Developers, product managers and QA get answers from your real code, Jira and Confluence.

Version 0.12.0 · October 12, 2026 · Release notes

All downloads
Adze finding a rounding bug in an invoice service: it searched the code, edited the file, ran the tests and summed up the change

Built for real work

For developers, and for the product managers, testers and delivery leads who work next to them.

Asks before it changes anything

Every file edit and command waits for your Allow or Deny. Edits show as diffs, commands with their output.

  • Deny with a note to tell the agent what to do instead, or always allow a tool for the chat.
  • Run several chats at once, in one project or many. The sidebar shows which are working and which need you.
  • A system notification tells you when a chat needs you while Adze is in the background.
A chat waiting for approval of a file edit, shown as a diff with Allow, Always allow and Deny, while another chat works in the sidebar

Reads your documents, not just your code

Attach PDFs, Word, PowerPoint and Excel files to a message, or paste and drop in screenshots, photos and designs.

  • Word and PowerPoint arrive with their headings and tables, every Excel sheet as a table, PDFs page by page.
  • A scanned PDF has no text, so the agent looks at its pages as pictures instead.
  • The agent can also open those formats in the project folder by itself.
A chat with a PDF, a Word document and an Excel workbook attached, and the brief the agent wrote from them

Documents you can hand on

Preview any Markdown document the agent writes or reads, with its tables, pictures and Mermaid diagrams.

  • The preview follows changes to the file as the agent works.
  • Export it to Word or PDF in one click.
A Markdown brief shown as a formatted document, with Word and PDF export buttons

Connected to Jira, Confluence and GitHubExperimental

Add an account once, then turn it on per project with that project's Jira key, Confluence space or repository.

  • Read only until you allow changes for a project, and then every change asks first.
  • Keys are encrypted with your system keychain and never reach a prompt or a transcript.
  • Any other app with an MCP server works too, hosted or run on your computer.
A project's connections: Jira and Confluence with a project and space key and changes allowed, and GitHub read only

Know what it costs

Estimated cost per chat, per project and per model, from Anthropic's list prices.

  • Daily and monthly budgets, and a calendar of the last 12 months.
  • A context ring and a prompt-cache countdown, with what a cache miss costs. Keep the cache for 1 hour for chats with long pauses.
  • How fast the provider answers: time to first token and output speed, per turn and over time.
The Usage page with the estimated cost, daily and monthly budgets, and a calendar of daily usage

See what the app sends

The Privacy page lists every domain the app talks to, what made each request, and how much data went each way. There is no Adze server in between.

  • Kept on your computer for 30 days, as sizes and addresses only. Export it whenever you like.
  • Anonymous usage statistics are sent only if you say yes, and never include your prompts, files or keys.
The Privacy page listing every domain the app connected to, what made each request, and how much data went each way

Works the way your role does

Pick Developer, Product Manager, Quality Analyst or Project / Delivery Manager, for the app or for one project.

  • Built-in skills for PRDs, user stories, backlog refinement, release notes, status updates and traceability.
  • Push stories to Jira and publish documents to Confluence.
  • Rewrite any persona's instructions to fit your team.
An empty chat for the Product Manager persona, offering to write a PRD, user stories or release notes

And a lot more

Anthropic API or AWS

Use an Anthropic API key, or Claude in Amazon Bedrock with your AWS sign-in, SSO included.

Your chats stay on your machine

Chats are saved as you go. Reopen one and the agent carries on with the full conversation.

SkillsExperimental

Agent Skills from your project or home folder. Type / to pick one, and see what each costs in context.

SubagentsExperimental

Hand read-only jobs, like exploring a codebase, to a cheaper model that reports back.

Slash commands

Type / for commands and skills: /compact, /context, /usage and more, completed as you type.

Help built in

A searchable guide inside the app, with info icons that open it at the right place.

Recovers by itself

Retries when the provider is busy or a reply stalls, and says plainly what went wrong when it can't.

Updates itself

New versions download in the background and install when you restart or quit. The Mac app is signed and notarized.

Readable replies

Replies show tables and highlighted code, and JSON in tool output comes indented and coloured.

Experimental features are off until you turn them on in Settings, under Experimental.

Install

macOS

  1. Download the .dmg. It's built for Apple Silicon Macs (M1 and later).
  2. Open it and drag Adze into Applications.
  3. Open Adze from Applications or Launchpad.

Requires macOS 12 or later.

Windows

  1. Download the .exe installer.
  2. Run it and pick where to install. No administrator rights needed.
  3. If SmartScreen says "Windows protected your PC", choose More info, then Run anyway.

Requires Windows 10 or later, 64-bit.

Before you start

You need an Anthropic API key, or an AWS account with access to Claude in Amazon Bedrock. The first launch walks you through picking one.

Your chats and settings are kept in the app's data folder, so updating to a new version keeps them.