Workflows/Build a local AI agent with Muse Glimmer 30B
Agents / WORKFLOW
Build a local AI agent with Muse Glimmer 30B
Run Meta’s Muse Glimmer 30B with llama.cpp, connect Hermes Agent using the settings on Meta’s GGUF card, and have it read two files and write a checked review.
Verified sourceSource checked 11/17/2026
LOCAL
Freshness 67 out of 100. Release age: 103 days old, Aug 10, 2026. Measures age, not quality.67/ 100Freshness103 days oldRelease age · Aug 10, 2026
Popularity 72 out of 100. 149,325 downloads in 30 days; 2,168 lifetime likes. Checked Nov 21, 2026. Measures adoption, not quality.72/ 100Popularity149.3K downloads / 30d2.2K lifetime likes · Checked Nov 21, 2026
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.
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.
Scores as of Nov 21, 2026. Statistics checked Nov 21, 2026. Statistics are marked for refresh after 72 hours.Full scoring method →
Before you begin
Hardware
Estimated from Meta’s published file sizes and VRAM pairings plus runtime and context headroom. Muse Glimmer is a dense 30B model, so every parameter is used for every token: offloading layers to system RAM works but slows it down far more than it would a mixture-of-experts model. Not measured by YouRunAI.
Machine
Muse Glimmer 30B · Q4_K_M GGUF
Muse Glimmer 30B · Q4_K_XL GGUF
NVIDIA · 16 GB VRAM + 32 GB RAM
Runs with several layers in RAM; noticeably slow
Heavy offload; choose Q4_K_M instead
NVIDIA RTX 4090 · 24 GB VRAM
Recommended start (Meta’s 24 GB pairing); keep the starting context
Does not fit fully; offload a few layers or use Q4_K_M
NVIDIA RTX 5090 · 32 GB VRAM
Comfortable, with room for longer context
Recommended (Meta’s 32 GB pairing)
AMD Radeon · 24 GB VRAM
Fits; requires the HIP or Vulkan llama.cpp build on Windows (Linux tab here is NVIDIA only)
Partial offload; requires the HIP or Vulkan build
Intel Arc · 16 GB VRAM + 32 GB RAM
Partial offload through the Vulkan llama.cpp build; slow
Heavy offload; choose Q4_K_M instead
Apple Silicon · 24 GB unified memory
Too tight for this guide: the file alone is 16.8 GB
Cannot fit
Apple Silicon · 32 GB unified memory
Recommended Mac start; close other apps
Tight; use Q4_K_M at the starting context
Apple Silicon · 64 GB unified memory
Comfortable
Comfortable
CPU only · 32 GB RAM
Loads, but a dense 30B model on CPU is very slow; short tasks only
Needs 48 GB RAM or more
Both packages are Meta’s own GGUF files from meta-models/Muse-Glimmer-30B-GGUF. Meta also publishes an optional 1.6 GB DFlash drafter for speculative decoding and full-precision weights (about 64 GB of VRAM) in the base repository; this guide leaves both out to keep the setup simple.
LOCAL AGENT · HERMESSOURCE-LINKED SETUP
Build a local AI agent with Muse Glimmer 30B
Serve Meta’s GGUF in llama.cpp, point Hermes at it with the configuration from Meta’s card, and have the agent read two project files and write review.md. You approve each file action.
Choose a package Run its commands Check the result
01
BEFORE YOU BEGIN
Choose your package
Select the weights you want. The instructions, downloads and settings change with your choice.
The publisher has not provided a complete memory target for this task. Check its hardware guidance before starting.
CURRENT SETUP
Muse Glimmer 30B · Q4_K_M GGUF · official
Start at 65,536 tokens for this Hermes agent setup. The 16.8 GB file is only the weights; the runtime and context add more memory. Longer context needs more memory, so reduce it first if the server runs out. Estimate, not a YouRunAI measurement.
Freshness 67 out of 100. Release age: 103 days old, Aug 10, 2026. Measures age, not quality.67/ 100Freshness103 days oldRelease age · Aug 10, 2026
Popularity 72 out of 100. 149,325 downloads in 30 days; 2,168 lifetime likes. Checked Nov 21, 2026. Measures adoption, not quality.72/ 100Popularity149.3K downloads / 30d2.2K lifetime likes · Checked Nov 21, 2026
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.
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.
Set up Muse Glimmer 30B · Q4_K_M GGUF on your machine
Choose your computer first, then its GPU path. The instructions below update to match your selection.
CHOOSE YOUR SETUPWindows
01
Install llama.cpp with winget
Open PowerShell and install the official llama.cpp release with winget, then close and reopen PowerShell so llama-server is on your PATH. Meta’s card says Muse Glimmer needs llama.cpp build b10353 or newer; the second command prints your build number, and the third lists the GPUs llama.cpp can use. If only the CPU appears on an NVIDIA or AMD PC, download the CUDA or HIP Windows build from the official releases page instead and run the same commands from its folder.
In PowerShell, run this command and leave the window open. The first run downloads Meta’s 16.8 GB Muse-Glimmer-30B-KQuant-17GB-Q4_K_M.gguf file; later runs reuse it. The --no-mmproj flag skips the vision encoder to save memory, because this guide only works with text and code. The server listens only on this computer at http://127.0.0.1:8080, names the model muse-glimmer-30b, and starts at 65,536 tokens of context with one request slot (-np 1), so the whole context goes to you. The sampling settings are Meta’s recommended temperature 1.0, top_p 0.95 and top_k 64, and --jinja is required for the chat template embedded in the file. Wait until the log says the server is listening.
Open a second PowerShell window and run the official Hermes installer. Close and reopen PowerShell afterwards so the hermes command is available. This native installer does not need WSL.
These are the custom-provider settings shown on Meta’s GGUF card, with the server’s muse-glimmer-30b alias as the model name. Run them in the second PowerShell window. Hermes then sends every request to llama.cpp on this computer.
hermes config set model.provider custom
hermes config set model.base_url http://127.0.0.1:8080/v1
hermes config set model.default muse-glimmer-30b
05
Create two small project files
In the same PowerShell window, create a disposable folder with a launch note and a checklist. The checklist deliberately has no database backup before the migration; the agent should find that gap from the files.
PowerShell
New-Item -ItemType Directory -Force "$HOME\agent-review" | Out-Null
Set-Location "$HOME\agent-review"
@"
Launch day: Thursday 14 May
Owner: Priya
Database migration runs at 09:00 before the app deploy.
"@ | Set-Content notes.md
@"
1. Freeze merges
2. Run the database migration
3. Deploy the app
4. Announce the release
"@ | Set-Content checklist.md
06
Give Hermes a file-based task
From inside the agent-review folder, start Hermes and paste the task below. Hermes asks before running commands or writing files; read each request before approving it. Approve the file reads and the single write of review.md.
Setting
Value
Task to paste into Hermes
Read notes.md and checklist.md in this folder. Write review.md containing: the launch day and owner, a corrected numbered checklist that adds a verified database backup before the migration, and one sentence explaining the risk of migrating without a backup. Cite the file each fact came from. Do not change notes.md or checklist.md.
Open review.md. It should name Thursday 14 May and Priya from notes.md, add a backup step before “Run the database migration”, and explain the risk. Check that Hermes’ activity shows real file reads and one write; an answer only in the chat window is not a completed agent run. Type /exit to leave Hermes and press Ctrl+C in the server window when you are done.
03
SUCCESS CHECK · Muse Glimmer 30B · Q4_K_M GGUF
An agent-written review you can check
review.md names the launch day and owner from notes.md, adds a verified backup before the migration, and Hermes’ activity shows the file reads and one approved write.
Meta’s card names Hermes Agent as a tested harness and shows the configuration used here. YouRunAI has not run this exact task with this package; inspect the output rather than trusting the chat summary.
Check that the server was started with --jinja, the model name is muse-glimmer-30b, and you started Hermes inside the agent-review folder. A chat-only answer is not a passing run.
Hermes warns that the context is too small, or the server runs out of memory.
If memory allows, restart the server with a larger -c; if it ran out of memory, halve -c instead and keep the task to these two small files. Restart Hermes afterwards.
llama-server says it does not recognise the model architecture.
Meta’s card says builds b10344 and older cannot load Muse Glimmer. Update llama.cpp (winget upgrade llama.cpp, brew upgrade llama.cpp, or docker pull the server-cuda image), check llama-server --version shows b10353 or newer, and try again.
The installer finishes but hermes is not found.
Close and reopen the terminal. Follow any PATH or dependency message the installer printed.
REFERENCE LIBRARY
Sources and files
Original instructions, model files, and compatibility notes behind this setup.