Qwen Code 1.0 Open Sources Harness: Unified Event-Driven Agent Architecture

Qwen Code 1.0 has been released as open source, integrating the Harness capabilities previously developed for Qoder. Qoder serves as a ready-to-use AI coding workbench, while Qwen Code provides the underlying open tool engine for embedding agent capabilities into custom products and workflows.

Unified Event Mechanism

The most significant technical change in Qwen Code 1.0 is the introduction of a unified event mechanism. In traditional agent runtimes, user input, model inference, tool calls, and execution results are organized around separate entry points and sessions. As external systems (IM, CI/CD pipelines) continuously connect, the challenge becomes how to uniformly receive, dispatch, and manage these streams.

Qwen Code 1.0 adopts events as the universal language connecting internal and external systems. External inputs are first converted into standard events via adapters, then routed to the appropriate task based on session identifiers. The session, combined with current state, decides the next step: invoke the model, execute a tool, or process new external feedback. Execution results are again transformed into events, feeding back into the task flow to drive the agent forward. This creates a continuously event-driven execution system, shifting from single-request triggers to a state management mechanism capable of handling diverse external inputs and advancing tasks persistently.

Scalable Agent System: From In-Process Bus to Distributed Architecture

With IM and automation pipelines feeding agents, the system must handle many concurrent tasks and feedback streams, demanding higher throughput, resource management, and extensibility. Qwen Code 1.0 evolves from an in-process memory event bus toward a distributed architecture along the unified event model.

The evolution proceeds through layered scaling of ingress, scheduling, and execution, moving from "one process, multiple sessions" to "multiple nodes, more sessions." A key design is "hand-brain separation": inference orchestration (brain) and tool execution (hand) are managed independently, allowing different processing units to collaborate around events. Qwen Code 1.0 also unifies management of models, sandboxes, and MCP (Model Context Protocol) infrastructure. These designs enable flexible scaling when adding more models, tools, and concurrent tasks while improving resource utilization efficiency.

Promotional Offer

From November 11 to November 24, users who log into Qwen Code with a Qoder account can claim a free Qwen4 trial quota valid for 30 days after redemption.