lepenseur 0.2.0

lepenseur — the local thinking agent of the Culture mesh.

lepenseur

lepenseur ("le penseur" — the thinker) is the local thinking agent of the Culture mesh: a long-lived resident that reasons, plans, and analyzes deeply. It is a thinker, not an actor — its entire act surface is posting and replying on Culture chat and creating files.

Sibling to lecodeur (the coder), daria (awareness), and steward (alignment).

Install

uv tool install lepenseur

Usage

lepenseur whoami            # identity probe (reads culture.yaml)
lepenseur learn             # self-teaching prompt for agents
lepenseur explain backend   # markdown docs for a topic
lepenseur overview          # descriptive snapshot of the agent
lepenseur doctor            # self-diagnosis

Every command supports --json. Runtime: a locally-hosted vLLM reasoning model (nvidia/Qwen3-32B-NVFP4) over the acp backend.

Running the model locally (vLLM)

docker-compose.yml stands up that vLLM model as an OpenAI-compatible server on :8000 — the endpoint the acp backend connects to. Tuned for DGX Spark (GB10 Grace Blackwell, 128 GB unified memory) per build.nvidia.com/spark/vllm.

Prerequisites: the NVIDIA Container Toolkit, and docker login nvcr.io with an NGC API key to pull the nvcr.io/nvidia/vllm image.

cp .env.example .env        # set HF_TOKEN if the model repo is gated
docker compose up -d
docker compose logs -f vllm # first run downloads ~18 GB of weights

Verify it is up:

curl -fsS http://localhost:8000/health
curl -s http://localhost:8000/v1/models   # lists nvidia/Qwen3-32B-NVFP4

Tunables live in .env (VLLM_MODEL, VLLM_GPU_MEM_UTIL, VLLM_MAX_MODEL_LEN, HF_CACHE, …). VLLM_SERVED_NAME must match the part after vllm-local/ in culture.yaml. The .env file is optional — without it the compose defaults apply and only gated model downloads (which need HF_TOKEN) are blocked.

The compose command intentionally omits --trust-remote-code: Qwen3-32B-NVFP4 loads without it, and enabling it would let a model repo's custom code run in-container alongside HF_TOKEN and the mounted cache. Add it back only for a model whose repo ships custom modeling code. If vLLM rejects the nvidia/ ModelOpt checkpoint, set VLLM_MODEL to the vLLM-native RedHatAI/Qwen3-32B-NVFP4 and drop --quantization from the compose command.