What Is MyAgents?
MyAgents's own GitHub repository describes it as "an elegant, easy-to-use Agent desktop app — an all-in-one Agent workbench and task center," distinct from a single coding-agent CLI or a cloud-hosted agent-building SaaS.
- Core idea: one desktop install that gives a person a GUI to work with agents directly, while also turning their computer into an execution node other agents (including agents on other MyAgents installs) can call on, per the README
- Ships as a desktop app for macOS 13+ (Apple Silicon and Intel) and Windows 10+; the repository also contains a
build_linux.shscript, but no Linux build is published on the Releases page, so Linux is not an officially supported platform as of this check - Maturity: the repository has roughly 0+ GitHub stars and is on release v0.5.13 as of this check
- What it is not: MyAgents does not train or run any AI model itself — it is a workbench and scheduler sitting in front of cloud-subscription agent runtimes (Claude Agent SDK, Codex) and API-key providers, comparable in that respect to how Askimo launches an external coding-agent CLI rather than reimplementing one
How MyAgents Works: Runtimes, Tasks, and AgentNet
Per the bundled in-app documentation and the README, MyAgents's features fall into four groups.
- Providers and runtimes — a user signs in to a supported AI subscription (Anthropic/Claude Code credentials, a managed Codex subscription, Grok/xAI OAuth, or Google Antigravity OAuth) or configures a generic API-key provider for an Anthropic-, OpenAI Responses-, or Chat Completions-style endpoint; each Agent then runs on a chosen runtime — the built-in Claude Agent SDK, the built-in "DeepSeek Harness" (works with whichever provider's models are active, not only DeepSeek), or a system-installed Claude Code CLI / Codex CLI
- Agent workspaces and Skills — each Agent keeps its own workspace folder, model/runtime choice, and
SKILL.md-based Skills and tools, and can hold multiple sessions at once; a user references files with@and invokes Skills with/inside a conversation - Task system — an Agent (or a user, from the task center) can create a Task that keeps its own objective, state, and run history, triggered once at a set time, on a fixed interval, or by a Cron expression; for work that only needs AI when something actually changes, a Task can first run a local script on a schedule and wake the Agent only once that script detects a condition (a finished build, a new file, a changed service status)
- Channel IM connections — built-in integrations for Feishu, DingTalk, and Telegram, with plugins documented for more platforms, so a user can message an Agent from an everyday chat app instead of opening MyAgents
- AgentNet — Agents made available under the same account on different computers can hand off work, files, and context to each other; the README states task content and results travel between devices end-to-end encrypted, and no server operated by MyAgents is required to make this work
- Space collaboration — organizes people and Agents around shared Issues and goals, for coordinating a person's own multiple Agents or inviting outside collaborators toward a shared result
Get MyAgents
MyAgents installs with a single downloaded installer on both supported operating systems — a DMG on macOS and an EXE installer on Windows — per the v0.5.13 release assets, checked 2026-11-18.
Platform | How to get it |
|---|---|
| macOS (Apple Silicon or Intel) | DMG from GitHub Releases |
| Windows | EXE installer from GitHub Releases |
| Website | myagents.io |
| Source code (GitHub) | github.com/hAcKlyc/MyAgents |
The current version is 0.5.13, per github.com/hAcKlyc/MyAgents/releases, checked 2026-11-18. No separate privacy-policy, terms, or company page distinct from the main site was found as of this check — see Privacy and Data Handling below.
Privacy and Data Handling
MyAgents's own marketing copy and README describe it as "local-first": the app and a user's files, environments, and task history are meant to stay on the user's own computers, with no MyAgents-operated server required.
- The README states plainly that "no server of your own is needed" and that data, environments, and Agent execution run on the computers a user already owns
- AgentNet, which hands work between a user's own devices, is described as transmitting task content and results end-to-end encrypted — this review could not independently verify the encryption implementation from the documentation checked
- MyAgents itself runs no model inference, so the contents of a conversation, a file an Agent reads, or a Task's results are visible to whichever cloud provider that session is configured to use (Anthropic, OpenAI/Codex, xAI/Grok, or Google Antigravity) under that provider's own privacy policy — MyAgents's local-first claims apply to the app and its own data, not to what a chosen cloud provider does with a request
- Channel IM integrations (Feishu, DingTalk, Telegram) mean messages sent to or from an Agent pass through that messaging platform as well, under its own terms
- No dedicated privacy-policy page, terms page, or company/about page distinct from the main single-page site was found as of this check, including after reading the site's raw HTML and JSON-LD directly
MyAgents License and Pricing
MyAgents the app is free; a separate commercial *license* exists only for organizations redistributing its code outside the terms of the AGPL.
- License: AGPL-3.0, confirmed via the GitHub repository's license field and the LICENSE file text
- Using the app: free, with no priced tier found anywhere checked — the site's own JSON-LD data states a price of "0" USD and marks the product as free to access
- A separate commercial software license is documented in LICENSING.md and COMMERCIAL-LICENSING.md for organizations that do not want, or cannot, comply with the AGPL when embedding MyAgents-owned components in a proprietary product, OEM/white-label distribution, or a proprietary hosted SaaS built on it — this is a licensing path for redistributors, not a paid tier an ordinary user of the app needs; terms are negotiated directly (contact listed as myagents.io@gmail.com)
- Running agent sessions still costs whatever the chosen AI subscription or API key costs outside MyAgents itself — MyAgents does not meter or bill for that usage
Who Should Use MyAgents?
Whether MyAgents fits depends mainly on whether you already pay for a cloud coding-agent subscription and want one desktop app to schedule and coordinate it, rather than wanting a local-inference tool.
Use MyAgents if
- You already use (or plan to use) Claude Code, Codex, or another supported cloud AI subscription and want a GUI workbench for it, with multiple agent workspaces and sessions at once
- You want recurring agent work to run on a schedule, or only once a local script detects a real change, instead of manually re-prompting an agent
- You run more than one computer and want an agent on one to hand off work and context to an agent on another without setting up your own server
- You want to message an agent from Feishu, DingTalk, or Telegram instead of keeping a dedicated app window open
- You want a free, AGPL-3.0, no-cost desktop app to evaluate before committing to any paid AI subscription behind it
Wait, or look elsewhere, if
- You need the agent runtime to run a model entirely on your own hardware — MyAgents runs no local model inference anywhere; see Askimo or Goose below for tools with a local-model path
- You need a stated company, support contact, or legal jurisdiction behind the product — none is published as of this check
- You need an officially supported Linux build today — only macOS and Windows installers are published, despite a Linux build script existing in the source repository
- You want end-to-end encryption claims (AgentNet) independently audited before relying on them for sensitive cross-device work — this review could not verify the implementation beyond the project's own description
MyAgents vs. Other Desktop Agent Tools
MyAgents is one of several desktop tools built around running an AI agent rather than a plain chat window. Here is how it compares to others in the same general agent-workbench and orchestration segment — see the Local AI Directory for the full catalog.
- Askimo — an AGPL-3.0, open-source desktop chat client for macOS, Windows, and Linux that also launches an external coding-agent CLI (Claude Code, Codex CLI, Antigravity CLI) as a subprocess, but is built first as a multi-provider chat client with RAG over a user's own files and a built-in local model option; MyAgents has no chat-client or RAG layer and no built-in local model, but adds task scheduling, IM-channel connections, and cross-device AgentNet that Askimo does not have. See the Askimo review.
- Goose — an open-source, Linux Foundation-governed AI agent for coding, automation, and workflows that markets itself as local-first and supports a documented local-model path alongside cloud providers; unlike MyAgents, Goose does not ship a built-in task scheduler, IM-channel integrations, or a cross-device agent network. See the Goose review.
- CrewAI — an open-source Python framework for defining and orchestrating teams of AI agents in code, aimed at developers building multi-agent systems rather than at a GUI for day-to-day agent use; MyAgents is a desktop GUI with no coding required, at the cost of the fine-grained programmatic control CrewAI offers. See the CrewAI review.
Frequently Asked Questions
Does MyAgents run AI models locally?
No. Every runtime and provider documented for MyAgents (Claude Agent SDK, Codex, DeepSeek Harness, and the Anthropic/OpenAI/Grok/Antigravity providers) calls a cloud subscription or API key. No Ollama, LM Studio, or other self-hosted model backend is named anywhere checked.
Is MyAgents free to use?
Yes, per the site's own JSON-LD pricing data (price "0" USD) and no paid app tier found anywhere checked. A separate commercial software license exists only for organizations redistributing MyAgents-owned code outside the AGPL, which is unrelated to what an ordinary user pays to use the app.
Does MyAgents support Linux?
Not officially as of this review. The v0.5.13 GitHub release ships only a macOS DMG and a Windows EXE installer; a build_linux.sh script exists in the source repository, but no Linux build artifact is published.
What is AgentNet?
AgentNet is MyAgents's feature for linking a user's own computers, under the same account, so agents can hand off work, files, and context to each other. The README describes this traffic as end-to-end encrypted and states no MyAgents-operated server is required; this review could not independently verify the encryption implementation.
Is MyAgents the same as DeepSeek Harness?
No. "DeepSeek Harness" is one of three Agent runtimes built into MyAgents, alongside the Claude Agent SDK and a managed Codex runtime. Despite the name, the bundled documentation states it works with whichever provider's models are currently active, not only DeepSeek models.
