Yao Agents

Yao lets you run and manage agents across all your machines from a single interface, which is an interesting idea for distributing work. The task board and multi-device support are well-executed, and you cannot beat the price. The problem is what the agents can actually do; without file system access, terminal execution, or browser control, their usefulness for software engineering tasks is severely limited. They can read from a knowledge base and talk to an API, but they cannot build or ship software.

Pick Yao Agents if you want a free, self-hosted dashboard to orchestrate non-coding tasks across your personal hardware. Pick anything else, like Aider or OpenDevin, if your goal is to automate software development.

Yao Agents provides no terminal, browser, or file system access. The term "agentic" appears on the homepage, but the platform lacks a sandbox. This means any agent with code execution capabilities would run directly on the host device, posing a significant security risk. The platform is a user interface for coordinating work across machines, not a secure environment for autonomous tools to perform that work.

Yao Agents offers a task board and an API to track agent activities across multiple devices. It is free and self-hosted. It excels at organizing and visualizing tasks for existing, trusted agents you bring to the system.

Yao Agents is documented as a self-hosted platform for running and managing agents across multiple user-provided devices [1]. The architecture is described as a set of isolated workspaces with a shared knowledge base, managed via a task board and accessible through an Open API [1]. The platform itself does not support user-provided models, and there is no documentation on the underlying models or agent capabilities for core tasks like file editing, terminal execution, or web browsing.

The system is presented as a harness, but what it is harnessing is not specified. Without information on agent architecture, model versions, or basic software development capabilities, the platform's utility for engineering tasks cannot be assessed. It appears to be a management interface for an un-specified, proprietary agent service.

Yao Agents is a well-designed product for a clear use case: running agents across a personal fleet of devices. The shipping cadence is consistent, and the multi-platform support is a real distribution advantage. But a self-hosted, proprietary, and entirely free product is a red flag. This isn't a business model; it's a user acquisition strategy funded by venture capital. The burn rate to support this must be significant.

When the runway gets short, expect a pivot to an enterprise offering or a pricing model that alienates the early adopters. The most likely exit is an acqui-hire by a device manufacturer or cloud provider looking for a multi-device orchestration layer, perhaps someone like Canonical or even a hardware player. Position: explore for personal productivity, but avoid for any critical business workflow.

The product is a multi-device, self-hosted agent harness. It is free to use, which means the only cost is the operational overhead of managing our own infrastructure. This is a personal productivity tool, not a platform for team-wide deployment.

There are no enterprise features. The specification lacks SSO, SCIM, audit logs, or a data retention policy. The architecture, running agents on user devices, also presents significant security and management challenges at our scale. Onboarding sixty developers would mean managing sixty separate, un-auditable instances. The lack of sandboxing for agent execution is a non-starter.

Yao Agents lets you run agents across your devices from a central task board, which is an interesting architecture. It's self-hosted and free, so the cost is just my own hardware. The multi-device, multi-workspace approach is clever for managing concurrent tasks.

But that's where my interest ends. It's proprietary source, so I can't read it, fork it, or fix it. There's no BYOK, no local model support, and no mention of what 'DeepSeek Harness' actually is or what models it uses. It's a closed system that I can host, but not own.