Agent Mate - Smart Clouds

AgentMate

A control center for your AI coding agents. Manage every CLI on your machine, track tokens and cost, watch GitHub activity and Actions, and keep a desktop pet that reports usage and pipeline failures.

Download

Get the latest AgentMate

Version 1.39.0 — pick your platform. Links always point at the newest release.

Windows Installer EXE Download macOS Apple Silicon DMG Download macOS Intel DMG Download Linux AppImage AppImage Download Android Companion app APK Download

Inside the app

One place for every AI coding tool you run

**Dashboard** CLIs, GitHub, usage, and system health at a glance.

**GitHub activity & Actions** Contribution heatmap, Actions trends, and the pet badge on the activity card.

**Token Usage** Tokens, cost, and quota across every provider.

**AI Pet** Click the companion for a token report. It also tells you when a pipeline fails.

**Projects** Bootstrap and launch the repos you work on.

**AI CLI Manager** Detect, install, and update coding CLIs.

What's new

GitHub charts, Actions, and a desktop pet

The dashboard now watches the repos you already work in. Contribution heatmaps, Actions trends, a single notifications list, and an optional companion that tells you when a pipeline fails.

Git charts

A GitHub contribution heatmap on the dashboard shows this week and this year for the account you connected. Activity sits next to system health, so you can see whether the repos are moving without leaving AgentMate.

GitHub Actions

Watch passed and failed workflow runs across the repos you connected. A trend chart on the dashboard makes a failing pipeline obvious before you open GitHub, and each project still has git tools for status, branches, tags, and Actions.

Notifications

GitHub Actions failures from the repos you connected land in one list. Settings has a Notifications tab, and the toast history stays searchable.

AI Pet

An optional desktop companion. Click it for a token report, double-click to bring AgentMate to the front, and let it tell you when a pipeline fails or the network drops. Right-click to snooze it, stop it wandering, or add your own GIF, PNG, or WebP character.

Key features

Everything your agent stack needs in one app

Dashboard

See every AI CLI, GitHub activity, GitHub Actions, and system health (CPU, GPU, memory, network) on one screen. Rearrange the cards, and see which apps are using the most resources.

Token usage & cost

Track tokens, cost, and rate-limit quotas across 60+ providers. Local-log scanning for Claude Code and Codex needs no API key, plus floating glass widgets that stay on top.

Projects

Bootstrap a repo, keep notes, run git (status, branches, tags, GitHub Actions), and launch each project's run command from its card without hunting through terminals.

AI CLI Manager

Detect every AI coding CLI on the machine and install missing ones with one click. Claude Code, Codex, Cursor, Gemini, Grok, and more stay in sync.

Prompt builder & history

Describe what you want and AgentMate structures it into a professional prompt for the agent you choose. Every prompt stays searchable in history.

Skills, MCP & remote

Install agent skills and MCP servers from configurable marketplaces. Check a skill before an agent reads it. Remote-control another AgentMate over your local network, including from the companion mobile app.

Skill security

Check a skill before an agent reads it

A skill is instructions an agent acts on, so its text is as powerful as code — and most skills come from a repo you have never read. AgentMate can scan one before you trust it: a static check over 14 risk categories, an optional deep review by an installed agent CLI, and a saved history of every run.

Where you run a check

Use the shield on any skill card — Directory, Featured, your repositories, or skills already installed. Skills → Security can take a folder path or a GitHub URL and check one skill or all of them, even if nothing is installed yet. A project page also lists skills sitting in agent folders that AgentMate did not install.

A static scan on your machine

The rules match on intent, not a single keyword, so a legitimate deploy skill that mentions curl is not automatically flagged. Every finding includes the file, line number, and the offending line. The report is there to be read, not obeyed. Batch runs report progress per skill and stay out of the way.

Verdict and score

Findings become a 0–100 score and one of four verdicts: safe, caution, risky, or dangerous. Each rule is charged once no matter how many lines it matched. A single critical finding keeps the verdict off the reassuring end of the scale, whatever else the file looks like.

Optional deep review

Turn it on and an installed agent CLI — the default from Settings, or one you pick — gives a second opinion in plain language. The rules always decide the score. A review can add findings and pull the verdict down, but it can never pull it up. The static scan stays on your machine; every check is stored locally and included in backups.

What the scan looks for

Fourteen risk categories, from prompt injection and credential theft to hidden content and overbroad permissions.

Prompt injection Text that tries to override the agent's instructions or hide what it does.
Data exfiltration Sending your files, output, or environment to an outside endpoint.
Credential theft Reading keys, tokens, cookies, or wallet files that belong to you.
Privilege escalation Asking for admin rights, editing shell profiles, or installing services.
Supply chain Pulling packages from unpinned, private, or non-standard sources.
Remote code execution Downloading something and running it, or evaluating fetched text as code.
Anti-refusal Jailbreak framing that pushes the agent past its own safety rules.
System prompt leakage Asking the agent to reveal its system prompt, tools, or hidden context.
Memory poisoning Writing lasting instructions into memory files, rules, or agent settings.
Unsafe output handling Rendering or executing model output without escaping or review.
Dark-pattern payment funnel Upsells, urgency, or payment details pushed through the agent.
Hidden content Invisible characters or encoded blobs carrying instructions you cannot read.
Destructive actions Commands that delete, reset, or overwrite data without a way back.
Overbroad permissions Skipping approval prompts or claiming wildcard tool access.

What it brings together

Built for developers who run more than one agent

Every CLI in one place

Stop hopping between terminals and dashboards. AgentMate detects what you already have installed and keeps each tool updated from a single manager.

Cost you can actually see

Token spend and quotas across providers land in one view, so you know which agent is expensive before the bill surprises you at the end of the month.

Desktop today, mobile tomorrow

The Electron app is the daily driver on Windows, macOS, and Linux. The companion mobile app reaches another AgentMate on your network when you are away from the desk.

Need a hand getting set up?

Tell us which AI coding CLIs you already use. We will show you how AgentMate ties them together and what your token dashboard looks like on day one.

Talk to us