Every review of agentic platforms starts the same way. It lists some prompts and massive frameworks to build and connect. The end result is typically a collection of disconnected agents.
This all changed when I switched to Moxby, a free tool that orchestrated all the agents I built by just talking to it.
This is the review I wished existed when I first started automating my tasks.
What Moxby Actually Is
Moxby is a desktop AI agent application for macOS and Windows. It runs on your machine. You connect your own AI subscription, whether that is Claude, ChatGPT, Gemini, or another provider, and then create specialized agents that work autonomously on your computer.
Each agent gets a SOUL.md configuration file. This is a plain text file that defines the agent’s role, its tone, its operating rules, and what it is responsible for. You edit the file to change how the agent behaves. You do not need to code anything. Clone a SOUL.md, modify it, and you have a specialized variant of that agent.
Agents run on a heartbeat schedule. You set the interval. The agent wakes up, checks its task board, processes whatever work is queued, writes outputs to the docs layer, and goes dormant until the next cycle.
The platform includes a full Kanban task board, a docs and pages system, browser automation, and four ways to teach agents new skills. You can point Moxby at a document or URL and it will read and absorb the content. You can use Learn a Web App to teach it any web tool by demonstration. You can install agents from the Moxby marketplace. Or you can import skill files directly.
Here is what matters about the architecture. Moxby is a coordination layer. It does not run its own AI models. It connects to your AI provider using your own API keys. Your data moves from your machine to your AI provider, and that is the full extent of it. Moxby as a platform does not insert itself as an additional data handler. That matters for certain use cases and is worth knowing before you evaluate it against cloud-based alternatives.
The Thing Nobody Gets Right About Moxby
Every review of Moxby starts the same way. It lists the features. Agent framework. Task board. Browser automation. Skills. Read and Learn. Learn a Web App. That is the wrong place to start.
The agents are not the product. The Initiative operating system that manages the agents is the product. Until you understand that, everything else looks like a collection of disconnected tools.
This is the review I wished existed when I first evaluated Moxby. It starts with the Initiative system because that is where Moxby is different. Everything else follows from there.
How the Initiative System Works
An Initiative is a managed project workspace. When you create one, Moxby creates it with three things automatically assigned.
A PM agent. This is a specialized agent whose entire job is to own the Initiative. The PM reads your objective, breaks it into a task board, defines the phases of work, recruits specialist agents to handle each phase, sets heartbeat schedules for those agents, monitors their progress, flags blockers, and generates weekly status reports. You are not managing the details. You are setting the objective and approving the work.
An Eval agent. This is a separate audit agent that monitors the PM’s own performance against the success metrics you defined. If the Initiative is falling behind on its targets, the Eval agent surfaces that independently of what the PM reports. You get two views: what the PM says is happening, and what the Eval agent says about whether the PM is actually hitting the goals.
A scoped workspace. The task board, docs, and agent configurations for this Initiative are contained within it. Nothing leaks into other Initiatives unless you explicitly move work between them.
The PM agent manages dependencies between tasks, not just the tasks themselves. A task does not move to done until the downstream step confirms it received the right input. If a specialist agent stalls, produces a bad output, or hits a blocker, the PM routes it to you. The agent does not silently proceed on bad data and surface a mess later.
This is the Initiative model: structured coordination with an audit trail, where the PM owns execution and the Eval agent keeps the PM honest.
A Real Security Team Workflow Example
Here is how this plays out in a security team context, because the abstract description only gets you so far.
Your Initiative objective: Automate the documentation pipeline for your Q2 VAPT engagements.
The PM reads that, creates a task board, and breaks it into five phases.
Phase 1 is reconnaissance. A specialist agent pulls scope data from your project tracker or ticketing system and stages it for the next step. This is the input that the rest of the pipeline runs on.
Phase 2 is vulnerability scoring. A specialist agent takes the scan output from Phase 1 and formats each finding against your severity taxonomy. CRITICAL, HIGH, MEDIUM, LOW. It applies your defined scoring rules automatically instead of doing it manually for every finding.
Phase 3 is report drafting. A specialist agent takes the scored findings from Phase 2 and generates a draft vulnerability writeup for each one. It follows your report template from the docs layer, references the scope document, and produces a structured first draft.
Phase 4 is QA review. A specialist agent reviews each draft against your deliverable standard. It checks that every required section is present, that the severity rating matches the scoring data, that the remediation advice aligns with your template. Drafts that pass move forward. Drafts that fail get sent back for revision.
Phase 5 is delivery. The PM routes the finalized, QA-approved report to wherever it needs to go.
The PM tracks every handoff between phases. If Phase 2 produces malformed data that Phase 3 cannot process, Phase 3 stalls and the PM flags it before you even notice. You decide what happens next. The agent does not silently ship a broken output.
Every phase is a tracked task. Every handoff is logged. The Eval agent compares actual throughput against your stated targets every week and reports back without you asking. You spend your time on technical decisions, not project coordination.
Sub-Agents and How Specialization Works
Moxby’s sub-agents are configured instances, not forked copies of a single prompt. Each has its own SOUL.md, its own tool access, and its own context window. You can run multiple sub-agents simultaneously, each handling a different part of the same workflow.
A common pattern inside an Initiative is a writer agent drafting, a reviewer agent checking the draft, and a formatter agent turning the reviewed output into the final deliverable layout. Each agent knows its role from its SOUL.md. You define it once. The agents run from the task board without needing to be told what to do every time.
The platform gives you an agent roster view where you can see every active agent, its role, what it is currently working on, and what it last output. You can steer a running agent, send it a message, kill it, or reassign its tasks. The orchestrator view gives you oversight without requiring you to micromanage.
For teams, the Team tier adds shared agent access so multiple people can interact with the same agent configurations and Initiative history. For individuals, the Free and Pro tiers support multi-agent workflows fully.
Skills and Extensibility
Moxby has ways to teach an agent a new capability.
Read and Learn is available on the free plan. Point Moxby at a URL, document, or knowledge base and it extracts and absorbs the content. The agent can then reference that knowledge without you repeating it in every task.
Learn a Web App is available on the Pro plan. Open a web tool, perform the actions you want the agent to learn, and Moxby captures the underlying API traffic. It reverse-engineers the endpoints, extracts your authentication tokens, and generates a reusable skill package. Any agent can then replay those API calls with your credentials, even if the tool has no official integration.
This is relevant for security teams because VAPT tooling, ticketing systems, internal portals, and proprietary platforms rarely have official AI integrations. Learn a Web App gives you a path to connect them without waiting for vendor support.
Install from Marketplace is a one-click install for community-built agents and skills. The catalog had 23 agents as of April 2026.
Import Skill lets you load a skill file directly. Community skill files, exported configurations, or custom-built skills all load the same way.
The Task Board and Docs Layer
Moxby’s task board is a native Kanban system, not a third-party integration. Tasks support subtasks, priorities, due dates, tags, and recurrence rules. You can create separate task lists for different projects and switch between them.
The docs layer is also native. You create pages within an Initiative workspace, write in HTML-supported rich text with tables and code blocks, and both agents and humans can read and write to them. Pages organize into folders.
For a multi-agent workflow, the task board tracks what is being worked on and the docs layer holds the actual work products. Drafts, findings, analysis, meeting notes. Agents read from docs, write to docs, and route task completions. You audit the docs.
Pricing
Moxby does not bundle AI credits. You connect your own subscription.
The Free plan costs nothing. It includes the full agent framework, unlimited sub-agents, the task board, the docs system, browser automation, unlimited skills and memory, and Read and Learn upskilling. A single-user multi-agent workflow is fully functional at no cost.
The Pro plan costs $25 per seat per month. It adds Learn a Web App, Remote Access from a phone browser, Watch Me for desktop automation, and unlimited projects, tasks, docs, sheets, and canvases.
The Team plan costs $40 per seat per month. It adds centralized billing, document collaboration, shared agent access, shared chats, and dedicated support.
For a solo security practitioner, the Free tier is functional for building and running multi-agent workflows. Pro is worth it if you need to connect to external tools through Learn a Web App. Team pricing makes sense for small teams that need shared access to agent configurations and project history.
Pros and Cons
Here is the honest version.
What works well:
- The local-first architecture means the platform layer does not create its own data collection surface. You connect to your AI provider using your own API keys. Moxby runs on your machine and your data stays on your machine through the workflow.
- The Initiative system is a genuine project management layer, not just a task list. The PM agent coordinates specialists, tracks dependencies, flags blockers, and reports on progress autonomously.
- Sub-agents are configured, not scripted. You change behavior by editing the SOUL.md file, not by rewriting prompts.
- The Learn a Web App feature works for any web tool that has an API, regardless of whether it has an official integration.
- The free tier is genuinely functional, not a crippled trial.
Where it falls short:
- No Linux support. macOS and Windows only.
- You need your own AI provider subscription on top of Moxby’s cost.
- Multi-agent orchestration takes real configuration time. It is not a zero-effort setup.
- The marketplace is still small. There are 23 agents as of April 2026.
- No standalone mobile app. Pro includes a browser-based Remote Access workaround.
Frequently Asked Questions
What is the Initiative system in Moxby?
Initiative is Moxby’s project management layer. When you create an Initiative, Moxby assigns a PM agent to own the project and an Eval agent to monitor the PM’s performance. The PM breaks the objective into a task board, recruits specialist agents, sets their schedules, tracks handoffs, and flags blockers. The Eval agent audits whether the Initiative is hitting its defined targets.
Is Moxby free?
Yes. The Free tier includes the full agent framework, unlimited sub-agents, the task board, the docs system, browser automation, unlimited skills and memory, and Read and Learn upskilling. The Pro plan costs $25 per seat per month and the Team plan costs $40 per seat per month.
Does Moxby run AI models locally?
No. Moxby is a coordination and automation layer. It runs on your desktop and connects to AI providers through their APIs using your own API keys. The AI model processing happens on the provider’s infrastructure, not on your machine.
How does Moxby handle data and privacy?
Moxby runs on your desktop. It does not operate its own AI models or introduce an additional data handler between you and your AI provider. Your conversation data, task content, and document writes stay on your machine. You should still review your AI provider’s data policies for their side of that equation.
Can Moxby connect to tools without official integrations?
Yes. The Learn a Web App feature, available on the Pro plan, reverse-engineers a web tool’s API by capturing your browser traffic as you use it. It generates a reusable skill package that any agent can replay with your authentication. This works for most web tools that have an API backend.
What are the main limitations?
No Linux support, the requirement to bring your own AI subscription, and the learning investment needed to configure multi-agent workflows properly. The Initiative system is powerful but it requires a real upfront configuration investment.
Final Thoughts
Moxby is a desktop AI agent platform built around structured multi-agent orchestration. Its Initiative system gives you a PM agent that coordinates specialist agents across phased work, with an Eval agent tracking whether the whole system is hitting its targets. Its Learn a Web App feature lets you teach it to connect to external tools by demonstration. Its task board and docs layer provide a native project management and knowledge base system.
It is not a plug-and-play AI tool. Configuring multi-agent workflows takes real setup time, and the output quality depends heavily on how clearly you define your agent configurations. But for teams that need structured automation with an auditable PM layer, and for environments where data locality matters, it is one of the more deliberate options in this space.
The free tier is functional enough to build and test a real workflow before committing to a paid plan.
Moxby.com for download and more information.
