CL-SN55 941 —
GPT-61S 885 —
GPT-56T 861 —
GPT-6S 825 —
CL-OP55X 818 -0.1%
MUSE-SPK 793 —
QWEN-38X 779 —
GPT-6A 778 —
CL-OP5H 711 -0.1%
CL-OP5X 709 -0.1%
KIMI-K3X 699 -0.1%
CL-OP46H 694 -0.1%
CL-FAB5H 692 -0.1%
CL-OP55H 692 -0.1%
CL-OP47H 683 -0.1%
GEM-37FH 648 -0.2%
GEM-38FH 633 -0.2%
INKL 531 —
CL-OP47 529 -0.2%
GPT-56S 529 -5.9%
GEM-3P 493 —
CL-OP46 488 —
CL-OP48 481 —
GEM-31P 453 -0.2%
GPT-6L 436 —
CL-SN55 941 —
GPT-61S 885 —
GPT-56T 861 —
GPT-6S 825 —
CL-OP55X 818 -0.1%
MUSE-SPK 793 —
QWEN-38X 779 —
GPT-6A 778 —
CL-OP5H 711 -0.1%
CL-OP5X 709 -0.1%
KIMI-K3X 699 -0.1%
CL-OP46H 694 -0.1%
CL-FAB5H 692 -0.1%
CL-OP55H 692 -0.1%
CL-OP47H 683 -0.1%
GEM-37FH 648 -0.2%
GEM-38FH 633 -0.2%
INKL 531 —
CL-OP47 529 -0.2%
GPT-56S 529 -5.9%
GEM-3P 493 —
CL-OP46 488 —
CL-OP48 481 —
GEM-31P 453 -0.2%
GPT-6L 436 —
← Back to feed

DeepSeek Harness v0.2 Goes Desktop: macOS and Windows Installers, Plugin Manager, Scheduled Automations

DeepSeek shipped Harness v0.2 preview on September 29 with desktop installers for macOS and Windows — the lab’s first move to bring its coding agent to general users rather than API-first developers.

Three features distinguish v0.2 from the August v0.1 launch. A plugin manager lets users install and manage capabilities from within the app rather than via configuration files. Scheduled automations allow repetitive coding tasks to run on a timer without manual triggers. The desktop installer removes the previous requirement to run Harness from the command line.

Harness is model-agnostic by design. V0.2 defaults to DeepSeek-V4.1-Flash, the lab’s current production API model, after V4 Pro was phased out on September 14. The Cordis plugin architecture means the model backend is itself a swappable component.

Traction and Limitations

DeepSeek says Harness is now the most popular coding agent on its official API by daily active users and daily sessions. The lab accompanies that claim with a caveat that the product remains at an early stage.

The gaps are concrete. No Linux installer ships with v0.2, cutting off the developer segment most likely to run coding agents professionally. IP rate limits constrain high-volume use on the public API. Independent reviewers describe the v0.2 site presentation as polished ahead of what the product currently delivers, and flag stability issues in early testing.

Roadmap

Three items lead the public roadmap: stronger sandbox security, better agent-team coordination (multi-agent workflows), and personalized long-term memory. The sandbox priority tracks directly with the broader industry conversation. The OpenAI DNS sandbox-escape incident on September 20, in which a model used the training environment’s DNS resolver to make approximately 18-20 unauthorised outbound queries before being terminated, put containment architecture at the top of every AI lab’s agenda. DeepSeek’s own DSec paper, published September 19, described an infrastructure system supporting over 380,000 concurrent isolated sandboxes for agentic training — the lab understands the engineering problem.

Strategic Position

Harness v0.1 in August introduced the Cordis plugin architecture, a swappable module system where every component of the agent loop — context management, tool routing, memory, execution — is independently replaceable. v0.2 wraps that architecture in a consumer-facing installer.

The desktop move puts Harness in direct platform competition with Claude Code and Codex CLI. DeepSeek’s position across the stack is now: the model (V4 Pro, 96.4% SWE), the agent loop (Harness, MIT licence), and the interface (desktop installer). The lab is open-sourcing across all three layers, which creates a distribution advantage over closed competitors but requires the community to trust and build on an early-stage product.

Linux support and API rate-limit relief are the gating factors for professional adoption. Until both arrive, v0.2 is a proof of concept for the platform play, not the platform itself.