DeepSeek Harness just got another update. RC.8 landed two days after RC.7 — and it’s not just a bug-fix release. This one changes how the whole thing works.
The short version? DeepSeek Harness is becoming a central dispatch layer for AI agents. You give it a task, it figures out which agent should do what, and it calls in the right tools — including Claude Code and Codex — as needed. RC.8 makes that vision a lot more real.
Let me walk through what actually matters.
Multimodal input is finally here
You can now feed images into Harness alongside text. The DeepSeek model adapter supports native image requests, and core commands like /goal and /plan can handle mixed text-and-image input. The @ menu also lets you reference local files and previous session history.
So if you’re working on a UI mockup and want the agent to analyze it, you can just drop the image in. No more describing what you see in words.
Claude Code and Codex are now first-class teammates
This is the big one.
In RC.7, Codex and Claude Code subagent tasks were integrated into the Job Panel. You could see them running. But RC.8 takes it further: both can now be installed on demand as Profile Bundles. Codex also supports non-interactive permission modes and multiple named instances.
What does this mean in practice?
Harness is no longer just a tool that uses Claude Code or Codex. It’s a tool that orchestrates them. You define the workflow, Harness breaks down the task, and it spins up the right coding agent for each subtask. Claude Code handles one piece, Codex handles another, and Harness manages the whole thing.
I think this is the most interesting direction for agent frameworks right now. Instead of picking one agent and sticking with it, you build a system that can call the best tool for each job. Harness is betting on that future.
Tool calls got smarter
The web_search tool now supports concurrent queries. If you need to search multiple sources at once, it happens in parallel instead of one after another.
There’s also a new reportDelivery mechanism for subagents. When a subagent finishes a task, it can report back immediately and wake up the parent task. That means less waiting around. For long chains of tasks — like searching multiple sources and then summarizing — this makes a real difference.
Windows users get some love
The Windows PTY terminal now supports persistent PowerShell sessions, and it’s enabled by default in the Minimal preset. Installation and startup are also smoother: download size is smaller, and running dsh web locally now opens your browser automatically.
Honestly? That auto-browser thing is small but nice. I hate copying localhost URLs manually.
Bug fixes that actually matter
RC.8 squashes a few annoying issues:
Fixed image size problems that caused model requests to fail
Fixed a bug where cancelling streamed generation didn’t carry over displayed response prefixes to follow-up prompts
Fixed compatibility issues with some custom OpenAI-compatible gateways
The Python SDK also got updated to cover all four built-in Agent Presets, with dependencies for rg/glob search and MCP stdio tools.
What I actually think
Here’s the thing. RC.8 isn’t about any single feature. It’s about direction.
DeepSeek Harness is slowly turning itself into the operating system for AI agents. It doesn’t care which model you use or which coding agent you prefer. It just needs to be able to call them, manage them, and stitch their work together.
Two days between RC.7 and RC.8 is fast. Really fast. That tells me the team is shipping aggressively and iterating on real feedback.
I’m not 100% sure where this ends up. Will Harness become the default way people orchestrate multiple agents? Or will we see more specialized tools that do one thing really well? Hard to say.
But I’ll keep watching. This one’s worth paying attention to.
