-
Notifications
You must be signed in to change notification settings - Fork 0
ralph dgx.stub
id: ralph-dgx name: RalphDGX (DeepAgents Code CLI + Ralph Loop) repo_path: ~/Documents/sw30labs/repos/ralph-dgx remote_url: https://github.com/sw30labs/ralph-dgx.git primary_language: Python framework: DeepAgents + vLLM (OpenAI-compatible) category: agentic-frameworks stacks: [agentic, cli, dgx] ingested: 2026-07-22 last_commit_date: 2026-07-19
DeepAgents Code CLI (dcode) plus a Ralph/tenacity goal loop, adapted for NVIDIA DGX Spark (GB10 / Linux aarch64) against a local vLLM OpenAI-compatible server. Sibling of the (non-public) original ralph repo targeting Apple Silicon OMLX / Hy3-MLX — same install → patch → overlay pipeline, retargeted defaults: provider vllm, model Qwen/Qwen3-Coder-Next-FP8, endpoint http://127.0.0.1:8000/v1, Python 3.12.
A harness, not a fork: the ralphdgx CLI drives doctor (GPU + vLLM + install status) → install (create .venv py3.12, download upstream DeepAgents, apply patches, overlay) → test (model catalog + tiny chat completion) → run (launch dcode against local vLLM). ralph/ holds the goal-loop logic, ralphdgx/ the DGX glue; constraints.txt pins the patched dependency set; a dedicated script re-runs the LangGraph update path. Falls back to a cloud OPENAI_API_KEY when no local server is up.
DeepAgents (downloaded + patched at install time, not vendored), ripgrep for repo search, external vLLM server. pytest.ini + 3 test files.
No CI workflows. setup-and-run.sh bootstrap; Apache-2.0; TODO-ROADMAP.md tracks a personal roadmap derived from DeepAgents research.
- deepagent-azure-cli — same DeepAgents coding-agent family, wired to Azure OpenAI / local MLX instead of vLLM
- driftlab-dgx — same shared Desktop vLLM stack on the DGX host
- Tags: [python, cli, llm, vllm, cuda, local-inference, qwen, openai]