Workflows/Build a private AI agent with Qwen3.8-Flash-Next

Agents / WORKFLOW

Build a private AI agent with Qwen3.8-Flash-Next

Run the official high-memory package, connect Hermes to its private Ollama endpoint, and produce a checked project review.

Verified sourceSource checked 10/14/2026

LOCALRENTED GPU

Freshness 82 out of 100. Repository age: 52 days old, Aug 24, 2026. Measures age, not quality.

Popularity 88 out of 100. 2,290,500 downloads in 30 days; 6,235 lifetime likes. Checked Oct 15, 2026. Measures adoption, not quality.

What do these scores mean?

Two ways to compare models. Neither score measures answer quality, benchmark performance, or whether a model fits your machine.

Freshness · release age

A new release starts at 100. The score halves every 180 days: 100 × 2^(−age in days / 180). When the release date is unavailable, a repository date is a fallback—not proof of when the model was released. Missing dates stay unscored.

Check the date source

Popularity · adoption

Downloads over the past 30 days carry 80% of the score; lifetime likes carry 20%. Both use a logarithmic scale, capped at 10 million downloads and 100,000 likes. If likes are unavailable, downloads carry all the weight. Missing download statistics stay unscored.

round(80 × min(1, log10(1 + downloads) / 7) + 20 × min(1, log10(1 + likes) / 5))

Check the statistics source

Scores as of Oct 15, 2026. Statistics checked Oct 15, 2026. Statistics are marked for refresh after 72 hours.Full scoring method

Before you begin

Hardware

Full package footprint matters more than six active parameters.

Machine	Qwen3.8-Flash-Next · 125B MLX	Qwen3.8-Flash-Next · 125B Q4
NVIDIA GPU · 32 GB VRAM	No MLX NVIDIA route; use the Linux package on a larger server.	Below the 120 GB weight file; use a multi-GPU host.
NVIDIA GPUs · 160 GB aggregate VRAM	No MLX NVIDIA route; choose Q4 Linux package.	Capacity candidate; confirm Ollama distributes the model and leaves context headroom.
AMD GPUs · 160 GB aggregate VRAM	No MLX AMD route; choose a supported Linux GPU runtime.	Ollama backend compatibility and distribution need verification for these exact cards.
Intel GPUs · 160 GB aggregate VRAM	No MLX Intel route; choose the supported Mac package.	No verified multi-Arc distribution route for this package.
Apple Silicon Mac · 128 GB unified memory	Weight file is 105 GB; runtime headroom is tight and this is not a recommended start.	Q4 is the Linux route here; choose the MLX package on Mac.
Apple Silicon Mac · 192 GB unified memory	Capacity candidate with headroom for 8K context; measure memory pressure.	Choose the MLX package instead of Linux Q4 on Mac.
CPU only · 192 GB RAM	No MLX CPU route here; use a supported high-memory Mac.	May load through offload but agent steps are not a practical recommendation.
RunPod · NVIDIA GPUs · 160 GB aggregate VRAM	No MLX rental route; choose Q4 Linux.	Capacity candidate; validate model distribution and runtime before a long rental.

Package sizes are published; fit and speed ratings are estimates. Multi-GPU distribution must be confirmed on the chosen rental.

LARGE PRIVATE AGENT · HIGH-MEMORY MACHINESOURCE-LINKED SETUP

Run a 125B agent on your files.

The checkpoint is large even though only 6B parameters activate per token. Choose the MLX Mac or Q4 Linux route, test tool calling, and then use Hermes to create a source-checked review.

Choose a package Run its commands Check the result

BEFORE YOU BEGIN

Choose your package

Select the weights you want. The instructions, downloads and settings change with your choice.

Save your machine in My Hardware to get an automatic starting choice. You can always choose any package yourself.

CURRENT SETUP

Qwen3.8-Flash-Next · 125B MLX · official

Start at 8,192 tokens on 192 GB unified memory; the 105 GB file is only weights. Longer context consumes more memory.

Freshness unavailable. Date pending.

Popularity unavailable. Stats pending.

What do these scores mean?

Two ways to compare models. Neither score measures answer quality, benchmark performance, or whether a model fits your machine.

Freshness · release age

Check the date source

Popularity · adoption

round(80 × min(1, log10(1 + downloads) / 7) + 20 × min(1, log10(1 + likes) / 5))

Scores as of Oct 15, 2026. Statistics check pending. Statistics are marked for refresh after 72 hours. These statistics are due for a refresh.Full scoring method

View weight source

THE WORKFLOW

Set up Qwen3.8-Flash-Next · 125B MLX on your machine

Choose where you want to run it. The instructions below update to match your selection.

CHOOSE YOUR SETUPMac

Other operating systems

Windows: This package is MLX for Apple Silicon, not a Windows binary. Windows can use the Q4 package as a client to Linux instead. Compatibility source

Linux: This exact package is MLX for Apple Silicon; the separate official Q4 package is the Linux multi-GPU route. Compatibility source

Check the machine before downloading

Check 192 GB unified memory free for the full model, runtime and a short context. This is a conservative capacity starting point, not a measured speed claim.

Exact Ollama package

Install and start Ollama

Download Ollama for macOS and open it. In Terminal, confirm the app is running before the model download. Ollama manages the package files, so no manual model folder tree is needed.

Download Ollama for this operating system Official model package

Download the exact package and set a short context

In Terminal, pull qwen3.8-flash-next:125b-mlx. Start with 8,192 tokens when connecting the agent; the published maximum context is not a safe default for every machine. The download is large and can take time.

Package identity and disk size

Bash

ollama pull qwen3.8-flash-next:125b-mlx
ollama show qwen3.8-flash-next:125b-mlx

Check model tool calling

In Terminal, run the small read-only probe against qwen3.8-flash-next:125b-mlx. It must report a tool call; this only checks the model connection and is not the agent workflow outcome.

Download the connection check Ollama tool support and model package

Bash

python -m pip install openai
curl -fsSLo private-file-agent.py https://www.yourunai.com/workflow-files/private-file-agent.py
python private-file-agent.py 'qwen3.8-flash-next:125b-mlx' http://127.0.0.1:11434/v1

Install and connect Hermes Agent

Install Hermes on your own Windows, Mac or Linux computer. Choose Custom endpoint in hermes model, enter http://127.0.0.1:11434/v1 and exact model ID qwen3.8-flash-next:125b-mlx, and start with 8,192 context. For RunPod, tunnel port 11434 over SSH so the same loopback URL works privately; do not publish Ollama's unauthenticated API.

Official Hermes installers Connect an Ollama/custom endpoint

Have the agent produce a useful result

Create a disposable folder containing a short project brief and a checklist with one missing backup step. Ask Hermes to read both, write review.md with a corrected checklist and the missing risk, and cite which local file supplied each fact. Review its requested file access and write before approval. This real file-based outcome, not the earlier read_notes probe, is the agent task.

Hermes agent and model setup

Inspect the artifact and close the service

Open review.md and check that Hermes actually used file tools, included the launch day and backup risk, and produced a useful corrected checklist. If the exact model/package fails at tool use, mark it unverified rather than treating a plain answer as success. On a rental, save the artifact and stop the pod.

SUCCESS CHECK · Qwen3.8-Flash-Next · 125B MLX

A source-checked review.md from Hermes

Hermes reads the disposable project files and writes a corrected checklist in review.md with the launch day and missing backup risk. Inspect the actual file and tool activity; the preliminary read_notes probe alone does not pass.

The agent route is source-backed but has not been benchmarked by YouRunAI. The Linux multi-GPU fit is a capacity estimate, not a tested deployment.

Back to setup steps

When it doesn’t go to plan

The full package does not fit or answers slowly.

Use the exact package size, leave runtime and context headroom, begin at 8K–16K context, and choose a larger machine or a smaller model if it still offloads heavily.

The agent returns text but no tool call.

Check the selected Ollama package has tool support, update Ollama, then retry the bounded script. Do not promote plain chat to a working agent path.

The client cannot connect to the local server.

Keep Ollama running, check http://127.0.0.1:11434/v1, and use an SSH tunnel for a rental. Never expose the unauthenticated port publicly.

REFERENCE LIBRARY

Sources and files

Original instructions, model files, and compatibility notes behind this setup.

3 SOURCES

01Official Ollama package sizes and tool supportollama.com 02Qwen publisher modelhuggingface.co 03Hermes custom endpoint setuphermes-agent.nousresearch.com

The model behind this workflow

Qwen3.8-Flash-Next

Keep exploring

Text

SETUP GUIDE

Qwen3.5 · 9B private chat setup

Source-backed package and platform-route plan for private local chat with Qwen3.5 · 9B. Preserves the official Ollama 9B package, a smaller 4B family fallback, and a clearly labeled community uncensored GGUF option.

Freshness 41 out of 100. Release age: 230 days old, Feb 27, 2026. Measures age, not quality.

Build a private coding assistant with Qwen3.8 27B

Run Qwen3.8 27B with Ollama, connect it to Cline in VS Code, and make one reviewable code change in a tiny local project.

Freshness 76 out of 100. Repository age: 71 days old, Aug 5, 2026. Measures age, not quality.

Arena Open Source Web coding: rank 10 of 55. Ranked model: qwen3.8-27b. Original model ranking, not a benchmark of this package. Export dated Oct 13, 2026. Source checked Oct 15, 2026. Human preference ranking, not hardware suitability.+3

SOFTWAREOllama + Visual Studio Code + Cline + Python

Create images locally with FLUX.2 Klein

Generate images from text prompts in ComfyUI with the fast distilled or more adjustable base 4B model. Save your results and reuse the visual workflow.

Freshness 35 out of 100. Release age: 273 days old, Jan 15, 2026. Measures age, not quality.

SOFTWAREComfyUI

LOCALRENTED GPU

Verified source