TGraphX Packages VisionServeX

Package · Apache-2.0

VisionServeX

A local-first, license-aware Python framework for serving modern computer vision models through a unified Python API, CLI, and HTTP REST gateway. 39 commercial-safe core models. No data leaves your machine.

v3.11.0 Apache-2.0 Python 3.10+ Local inference License-aware

What is VisionServeX?

VisionServeX is an open-source Python framework for running modern computer vision models locally and exposing them through a stable HTTP API. It works as a local model gateway: start it once, call any supported model through one clean API, on your own hardware, with no data leaving your machine.

Every model in the registry carries an explicit license classification, an honest availability status, and a clear commercial posture. Blockers are documented — not hidden. If a model is unavailable, VisionServeX tells you exactly why and what to do next.

Install

bash
pip install 'visionservex[hf,rfdetr]'

The base package requires Python 3.10+ and PyTorch. Optional extras ([hf], [rfdetr], [server], [anomaly], [dino]) activate specific model families on demand.

Quickstart

bash
visionservex --version
visionservex getting-started        # personalized setup guide

# Detection (RF-DETR, no token needed)
visionservex detect rfdetr-small image.jpg

# Segmentation (SAM2.1, no token needed)
visionservex sam-family smoke-test sam2.1-hiera-small image.jpg

# HTTP gateway
visionservex serve                  # http://127.0.0.1:8080
curl -F "[email protected]" -F "model_id=rfdetr-small" \
     http://127.0.0.1:8080/detect | jq
python
from visionservex import VisionModel, VSX

# Direct inference (no server needed)
result = VisionModel("rfdetr-small").predict("image.jpg")
result.to_json()

# SAM2 segmentation
VSX.sam("sam2.1-hiera-small").segment("image.jpg", box=[10, 20, 200, 220])

# DINOv2 embedding (Apache-2.0, no token required)
VSX.dino("dinov2-base").embed("image.jpg")

Supported tasks and model families

Object detection

  • D-FINE (n/s/m/l/x variants)
  • RF-DETR (nano/small/medium/large)
  • Install: [hf] or [rfdetr]
  • Token: not required

Segmentation

  • SAM v1, SAM 2, SAM 2.1 — promptable, no token
  • RF-DETR-Seg — instance segmentation
  • SAM3 / SAM3.1 — BYOT gated (HF token + license)
  • MedSAM — medical, research only

Dense embedding

  • DINOv2 (s/b/l/g) — Apache-2.0, no token
  • CLIP, SigLIP2 — no token
  • DINOv3 — BYOT gated
  • Install: [hf]

Open-vocabulary detection

  • Grounding DINO (tiny, swin-b) — no token
  • OWL-ViT, OWLv2 — no token
  • Install: [hf]

Multi-task VLM

  • Florence-2 (base, large)
  • Detection, captioning, grounding, OCR
  • No token required
  • Install: [hf] in isolated env

Classification & depth

  • SwinV2, ConvNeXtV2, MaxViT — no token
  • depth-anything-small — no token
  • DINOv3 + CHMv2 DPT — BYOT depth head

Anomaly detection

  • PatchCore, PaDiM via anomalib
  • Install: [anomaly]
  • No token required

HTTP REST gateway

  • Full REST API for all tasks
  • Binds to 127.0.0.1:8080 by default
  • No data retained between requests
  • Install: [server]

License and commercial posture

Every model in VisionServeX's registry is explicitly classified into one of four categories:

Category Runs locally? Token needed? Commercial-safe?
Commercial-safe core (39 models) Yes No Yes (Apache-2.0 / MIT)
BYOT gated (SAM3, SAM3.1, DINOv3) Yes, after access Yes (HF token + upstream license) Depends on upstream license you accepted
Research / non-commercial Maybe Maybe No
External API only No local weights Provider key Depends on provider terms

The commercial-safe core includes SAM v1/2/2.1, DINOv2, RF-DETR family, Grounding DINO (open variants), Florence-2, CLIP, OWLv2, SigLIP2, SwinV2, ConvNeXtV2, and more — no token required; weights download from official upstream sources on demand.

BYOT models (SAM3, SAM3.1, DINOv3) are not automatically commercial-safe. Commercial use depends on the upstream Hugging Face license you accepted. VisionServeX provides the infrastructure; the license decision is yours. VisionServeX never redistributes gated model weights.

Bring Your Own Token (BYOT) for gated models

SAM3, SAM3.1, and DINOv3 are gated on Hugging Face. VisionServeX provides a clean BYOT path: supply your own HF token, accept the upstream license once, and weights stay in your local cache.

bash
# Connect your HF token
visionservex hf connect --token-env HF_TOKEN

# Check status (no download)
visionservex hf status
visionservex hf check-model facebook/sam3

# Pull after accepting the upstream license
visionservex model pull sam3-base --accept-upstream-license
visionservex model doctor sam3-base

Key design principles

Related reading and links