llama-atmosphere-agent: document the from-scratch workflow - #453
Merged
Merged
Conversation
The agent README assumed the reader already had the folder, a model and (for mode A) a fat jar. It now walks a new user through it: - "Getting started from scratch": git clone, a curl download of a tool-capable model (Qwen3-4B-Instruct-2507 from unsloth's GGUF repo), the start command for Linux/macOS and for Windows PowerShell, and the GPU classifier as an optional step. Examples use models/Qwen3-4B-... throughout instead of mixed model names. - Mode A says where the fat jars come from (GitHub release assets) and that the all-<os>-<arch> jars pick a GPU backend themselves (net.ladenthin.llama.backend forces one). - A warning box: with --allow-shell the model runs any command with the user's rights, without asking. - Limitations: one in-process agent per machine (fixed temp-dir native library name on Windows, shared GPU memory); several agents should share one server. - Root README: the same short path (clone, model, run, warning). - .gitignore: llama-atmosphere-agent/models/*.gguf, where the README downloads the model (the existing models/*.gguf rule only covers the root). Verified on Windows: the documented PowerShell command with the relative models\ path runs on the CPU jar, calls a tool and answers; the copied model is not shown by git status. Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01E2h8gXyyE5UeimkL9vQv9G
bernardladenthin
had a problem deploying
to
maven-central
September 21, 2026 22:10 — with 
GitHub Actions
Failure
GitHub Actions
Failure
bernardladenthin
had a problem deploying
to
maven-central
September 21, 2026 22:10 — with
GitHub Actions
Failure
GitHub Actions
Failure
bernardladenthin
had a problem deploying
to
startgate
September 21, 2026 22:10 — with
GitHub Actions
Error
GitHub Actions
Error
This branch had an error being deployed
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Why
The agent README assumed the reader already had the folder, a model and — for mode A — a fat jar. A user on a fresh machine with only JDK + Maven had to guess those steps.
What
llama-atmosphere-agent/README.mdgit clone --depth 1 …+cd java-llama.cpp/llama-atmosphere-agent(or the GitHub ZIP); the folder is self-contained,.mvn/jvm.configincluded.curldownload of Qwen3-4B-Instruct-2507 Q4_K_M (unsloth GGUF repo — the official Qwen repo returns 401 without login). All examples now usemodels/Qwen3-4B-…instead of a mix of model names.llama-<v>-jar-with-dependencies.jaris CPU (+ Metal),llama-<v>-all-<os>-<arch>-jar-with-dependencies.jarcarries every GPU backend and uses the first whose runtime loads (-Dnet.ladenthin.llama.backend=…forces one).--allow-shellruns any command with the user's rights, without asking.Failed to delete old native lib/No native library found; shared GPU memory); run one server and point several agents at it instead.Root
README.md: the same short path (clone, model, run, warning, PowerShell/GPU hint), links to the agent README..gitignore:llama-atmosphere-agent/models/*.gguf— the README downloads the model there, and the existingmodels/*.ggufrule only covers the root.Verified
Windows 11: the documented PowerShell command (
--model models\Qwen3-4B-Instruct-2507-Q4_K_M.gguf …, CPU jar) runs, callslsand answers; the model copied intomodels/does not show up ingit status. Model URL checked (HTTP 200).🤖 Generated with Claude Code
https://claude.ai/code/session_01E2h8gXyyE5UeimkL9vQv9G