Qwen 3.8 in DeepSeek Harness: Free, Local, One Catch
Read the Portuguese translation →
Author: Leon van Zyl · Watch on YouTube
Video authored by the channel shown, not by IANIA; all rights belong to the creator. Transcript and summary generated automatically; the original-language content is preserved.
Original language: English
Original video: Qwen 3.8 in DeepSeek Harness: Free, Local, One Catch · Channel: Leon van Zyl · 00:30 · Watch on YouTube
Topics: AI agents, Prompt and context engineering, Local and open-source models
Summary
Section titled “Summary”The speaker describes running a free model locally inside DeepSea Corners, with access limited to selected tools. They praise Queen 3.8 and say it requires about 16 GB of VRAM. They caution that local models are not well suited to sub-agents or the Ralph loop, and recommend creating presets that avoid wasting context on unusable tools. The speaker says their course link will be in the description.
Key points
Section titled “Key points”- A free model is running locally inside DeepSea Corners.
- The model is configured with only selected tools.
- The speaker is a fan of Queen 3.8.
- The model requires around 16 GB of VRAM.
- The speaker says local models are not very good at running sub-agents.
- The speaker advises against using local models for the Ralph loop.
- Custom presets can limit tool access.
- Avoid filling the context window with tools the agent cannot use.
- A course link will be included in the video description.
Chapters
Section titled “Chapters”| Start | Chapter | Summary |
|---|---|---|
| 00:00 | A free local model | The speaker introduces a free model running in DeepSea Corners with selected tools and mentions Queen 3.8’s VRAM needs. |
| 00:10 | Limits of local models | The speaker cautions against using local models for sub-agents or the Ralph loop. |
| 00:20 | Presets and course link | The speaker recommends limiting tools to protect the context window and mentions a course link in the description. |
Quotes
Section titled “Quotes”we have a free model running inside DeepSea Corners — 00:00
you’ll need round about 16 gig of VRAM to run this model. — 00:00
we really shouldn’t be running the Ralph loop using local models. — 00:00
You do not want to blow the context window with tools that the agent can never use. — 00:00
I’ll link to my course in the description of the video. — 00:30
Concepts and entities
Section titled “Concepts and entities”Qwen3.8, DeepSeek Harness, VRAM, Ralph loop, context window, Leon van Zyl
Full transcript (original language)
Section titled “Full transcript (original language)”00:00 Now, we have a free model running inside DeepSea Corners, and it’s optimized with only the tools that we want to give it access to. I’m personally a massive fan of Queen 3.8, and you’ll need round about 16 gig of VRAM to run this model. Local models are really not that good at running things like sub-agents, and we really shouldn’t be running the Ralph loop using local models. What we can do is create our very own presets. You do not want to blow the context window with tools that the agent can never use. I’ll
00:30 link to my course in the description of the video.