selfhost.directory

The index of self-hostable projects — ranked, categorized and served with live updates.
Own your data and stop giving it to big tech.

GitHub
locate-anything logo

Self-host locate-anything

v0.1.3

Web UI for NVIDIA LocateAnything-3B object detection

BetaOpen SourceApache-2.0
GitHub
Install
Updatedlast month
LanguageTypescript
Setup difficultyAdvanced
MobileResponsive
Stars
168
CPU
NVIDIA GPU with compute capability ≥ 8.0 (Ampere or newer)
RAM
~12GB+ VRAM recommended (8GB may work but can OOM on large images)
Storage
~6GB for model cache + space for uploaded images and SQLite database
Port
8000 (backend API (when exposed via override))
Docker
ghcr.io/gammahazard/locate-anything-backend
Install with Docker Compose
Home view — upload, prompt, task presets, decode modeDetection result — labeled reticle boxes show what was foundSystem view — GPU/health readout, GPU picker, and configurable backend URL

What is locate-anything?

locate-anything provides a sleek, mobile-friendly web interface wrapping NVIDIA's LocateAnything-3B vision-language model. You upload an image, enter a natural-language prompt, and receive bounding boxes for object detection, phrase grounding, text detection and OCR, document layout analysis, GUI element grounding, and pointing. The interface includes three speed/quality modes (fast, hybrid, slow), preserves search history for re-running, and offers a GPU preflight check. It requires an NVIDIA GPU with Ampere or newer architecture and roughly 12 GB of VRAM, deployed through Docker Compose with prebuilt images from GHCR or a local build. The underlying model carries a non-commercial license for academic and research use, while the interface code is Apache-2.0.

Should you self-host locate-anything?

locate-anything scores 42/100 on selfhost.directory (how scoring works) and ranks #95 of 125 in AI. It has 168 GitHub stars.

It shipped 4 stable releases in the past year. The latest, v0.1.3, came out on 2 June 2026.

Setup is advanced: plan for several services and some ongoing maintenance. An official Docker image is available. It is Apache-2.0-licensed and written mostly in Typescript.

Its closest alternative on selfhost.directory is ComfyUI (93/100, 143 741 stars). It scores higher overall, so compare the two before you pick. See locate-anything vs ComfyUI side by side.

Features

  • Document layout analysis
  • GPU preflight check
  • GPU selector in UI
  • GUI element grounding
  • Mock mode (no GPU)
  • Model caching
  • Non-commercial model license noticeModel license is NVIDIA non-commercial; UI code is Apache-2.0
  • OCR / text localization
  • Open-vocabulary object detection
  • Phrase grounding
  • Pointing
  • Remote GPU access
  • Search History
  • Speed/quality toggle

locate-anything README

How to install locate-anything

Install locate-anything with Docker Compose

Docker only (no GPU needed)
Environment variables
LA_MOCK, LA_MODEL_PATH, LA_DEFAULT_MODE, LA_DEVICE, LA_DATA_DIR, HF_HOME, HF_TOKEN, FRONTEND_PORT, BACKEND_PORT, LA_VERSION

mkdir la-app && cd la-app curl -O https://raw.githubusercontent.com/gammahazard/locate-anything/main/docker-compose.ghcr.yml mv docker-compose.ghcr.yml docker-compose.yml docker compose pull docker compose up

Pulls prebuilt images from GHCR; no source checkout or build needed. Pin version with LA_VERSION in .env.

How to configure locate-anything

  • BACKEND_PORT8000Host port for backend API (only used with expose-backend override)

    Host port for backend API (only used with expose-backend override)

  • FRONTEND_PORT9080Host port for the web UI

    Host port for the web UI

  • HF_HOME/hf_cacheHuggingFace cache dir (mounted as a volume so the ~6GB model downloads once)

    HuggingFace cache dir (mounted as a volume so the ~6GB model downloads once)

  • HF_TOKENOptional: needed only if the model repo is gated on HuggingFace

    Optional: needed only if the model repo is gated on HuggingFace

  • LA_DATA_DIR/dataWhere the backend stores the SQLite db + uploaded/annotated images (inside container)

    Where the backend stores the SQLite db + uploaded/annotated images (inside container)

  • LA_DEFAULT_MODEhybridDefault generation mode

    Default generation mode

    values: fast | hybrid | slow

  • LA_DEVICE0Which GPU to load the model on at startup (index from nvidia-smi)

    Which GPU to load the model on at startup (index from nvidia-smi)

  • LA_MOCK0Set to 1 to run without a GPU/model (serves deterministic fake boxes)

    Set to 1 to run without a GPU/model (serves deterministic fake boxes)

    values: 0 or 1

  • LA_MODEL_PATHnvidia/LocateAnything-3BHuggingFace model id or local path

    HuggingFace model id or local path

  • LA_VERSIONlatestOptional: pin a published image version for the prebuilt (GHCR) compose file

    Optional: pin a published image version for the prebuilt (GHCR) compose file

    values: e.g. 0.1.1

locate-anything releases & version history

Even more releases RSS

More in Computer Vision

Rate this project

Click to rate — no account needed.

0 comments

Posting as guest
Loading comments…