OTTCOUTURE Cannabis Vision OpenCore

The OTTCOPS analyzer from ottcouture.eu combines Teachable Machine vision models with multimodal LLMs to produce structured JSON outputs. All branding and rights remain with ottcouture.eu; feedback or model submissions are welcome via otcdmin@outlook.com, Instagram @ottcouture.eu, or Discord discord.gg/GMMSqePfPh.

Feature Highlights

• 🌿 FastAPI core with Analyzer UI, config deck, OTTO chat (/completions), and documented /tm-models* routes.
• 🧠 Vision LLM switchboard for OpenAI, Ollama, or LM Studio with system prompt presets, multiple profiles, and server-side persistence for the Analyzer, Streams, and OTTO.
• 🧪 Teachable Machine depot with ZIP uploads (TFJS: metadata.json/model.json/weights.bin or Keras: keras_model.h5 + labels.txt), registry, and default selection for the Analyzer.
• 🧵 Model routing: the frontend lets you pick a TM slot per run; the choice is persisted in app-settings.json for reuse.
• 🤖 OTTO grow chat – dedicated screen for cultivation questions with a defined system prompt.
• 📡 WiFi broadcast mode (mDNS/zeroconf) for hostnames like ottcolab.local across your LAN.
• 📝 Prompt templates including locally saved custom presets directly in the Analyzer.
• 🗂️ Batch analysis with /api/opencore/analyze-batch, tabs per image, and an overall report.
• 🛠️ Debug panel with request ID, model version, and timings (UI toggle + ?debug=1).
• 🔐 API token mode: custom base URL + token with code samples.
• 📤 Export bundle: JSON download, PDF report, and share links via /api/opencore/share + viewer (/share/<id>).
• 🧷 ML-only analysis mode: analysis_mode=ml returns Teachable Machine JSON without GPT.
• 🎥 Stream orchestration: snapshot/RTSP sources run as background jobs (5 s capture, 30 s batch) and produce automated reports.
• 🔄 Launch update check: on start the backend compares against github.com/ottco-dev/ottcops and offers an optional git pull.

Usage & Licensing

• OPENCORE Analyzer is free to use only for private individuals and developer testing.
• Cannabis Social Clubs (CSCs) and companies—startups, MSOs, or service providers—must obtain a commercial license directly from ottcouture.eu before using this in any production or revenue context.
• Licensing & partnerships: otcdmin@outlook.com, Instagram @ottcouture.eu, Discord discord.gg/GMMSqePfPh.

Installation (OTTCOUTURE style)

python -m venv .venv
source .venv/bin/activate
pip install -r requirements.txt

# optional if you want GPT calls
export OPENAI_API_KEY="sk-..."

# start the dev server
uvicorn app:app --reload

At startup the server automatically compares your checkout to https://github.com/ottco-dev/ottcops. If a newer commit exists you will be prompted in the console (“Update now?”). Reply y or yes to run git pull; any other response keeps the current version. Set OTTC_SKIP_UPDATE_CHECK=1 to bypass this (e.g., in CI).

1. Analyzer UI: http://localhost:8000/
2. OTTO grow chat: http://localhost:8000/completions
3. Config hub + TM depot: http://localhost:8000/config
4. Discord crew & support: discord.gg/GMMSqePfPh
5. Documentation (HTML): http://localhost:8000/doc/index.html

Configuration

Variable	Required	Default	Description
OPENAI_API_KEY	for OpenAI flow	–	Key for GPT-4.1 mini or your preferred vision model.
OPENAI_GPT_MODEL	optional	gpt-4.1-mini	LLM ID for cloud vision.
TEACHABLE_MODEL_PATH	optional	./models/teachable_model	Alternative path to a legacy Teachable model.

Provider/LLM configuration from the config hub is stored locally (localStorage.cannabisLLMConfig) and server-side via /api/settings/llm. Multiple profiles can be created, activated, or deleted through /api/settings/llm/profiles; selections appear in the Analyzer, Streams, and OTTO. Together with the default Teachable Machine model, values are written to app-settings.json so Analyzer, batch/stream endpoints, and OTTO share the same provider settings and survive restarts.

WiFi broadcast (ottcolab.local)

1. Install dependencies (zeroconf ships in requirements.txt; run pip install zeroconf if you reuse an existing environment).
2. Start uvicorn app:app --host 0.0.0.0 --port 8000 so the server is reachable on your LAN.
3. Open http://localhost:8000/config and scroll to “WiFi Broadcast & ottcolab.local”.
4. Set a hostname (we enforce .local) and confirm the port, then click “Enable broadcast”.
5. Devices on the same network can reach http://ottcolab.local:8000/. Feedback still welcome at otcdmin@outlook.com, Instagram @ottcouture.eu, or Discord.

Teachable Machine depot (/TM-models)

1. Export your Google Teachable Machine project as a TensorFlow bundle (metadata.json, model.json, weights.bin) or a Keras (.h5) bundle with keras_model.h5 and labels.txt.
2. Open http://localhost:8000/config and use the “OTTCOUTURE Teachable Machine Depot” section.
3. After upload the model is stored under /TM-models/<slug> and registered in TM-models/registry.json.
4. The server converts TFJS exports into a TensorFlow SavedModel (tensorflowjs is declared as a dependency). Missing converters or broken bundles return clear errors.
5. The list view lets you mark any model as “Default in Analyzer”. The default is mirrored in app-settings.json.
6. If no community model is selected, the Analyzer falls back to TEACHABLE_MODEL_PATH (OPENCORE reference).

Required files: either metadata.json, model.json, weights.bin or keras_model.h5 plus labels.txt. Missing parts are rejected during upload.

API routes

• GET / – Analyzer landing page with model selector
• GET /config – self-host configurator & TM depot
• GET /completions – OTTO grow chat UI
• POST /analyze – image + prompt + optional model_id + analysis_mode
• POST /api/opencore/analyze-ml – alias for ML-only calls (same as /analyze with analysis_mode=ml)
• POST /api/opencore/analyze-batch – multi-image analysis (FormData with files[])
• POST /api/opencore/share & GET /api/opencore/share/{id} – JSON share service (/share/{id} serves the viewer)
• POST /api/completions – OTTO chat endpoint (prompt in JSON body)
• GET/POST/DELETE /api/opencore/streams* – manage snapshot/video streams including trigger endpoint
• GET /tm-models – registry + default information
• POST /tm-models/upload – ZIP upload (file, model_type, display_name)
• POST /tm-models/default/{model_id} – set default model
• DELETE /tm-models/default – clear default model
• GET/POST/DELETE /api/settings/llm – persist provider/prompt configurations
• GET /network/status, POST /network/announce, DELETE /network/announce – mDNS controls

Documentation in /doc

All feature guides ship as static HTML pages served by FastAPI under /doc:

• doc/prompts.html – prompts & custom presets
• doc/batch.html – batch analysis with API examples
• doc/debug.html – debug panel
• doc/api_token_mode.html – professional mode
• doc/ui.html – UI extensions (drag & drop, theme, zoom, JSON fullscreen)
• doc/export.html – JSON/PDF/share export
• doc/home_automation.html – home-automation guide incl. curl, Python, Node-RED, Home Assistant
• doc/streams.html – video & snapshot streams with API calls
• doc/models.html – Teachable Machine (easy) and Label Studio/YOLO (pro) workflows
• doc/raspberry.html – Raspberry Pi mounting, camera setup, and edge scripting

Project structure

.
├── app.py                # FastAPI service + TM depot + WiFi broadcast + OTTO endpoint
├── static/
│   ├── index.html        # Analyzer UI with model selector
│   ├── completions.html  # OTTO grow chat
│   └── config.html       # Self-host + TM depot console
├── TM-models/            # Versioned Teachable Machine bundles (ZIP uploads)
│   ├── README.md
│   └── registry.json     # maintained at runtime
├── app-settings.json     # Default model & provider settings (created if missing)
├── requirements.txt
├── README.md
└── LICENSE

Feedback & rights

• Brand & rights: ottcouture.eu — this is released as OpenCore, but all branding remains with ottcouture.eu.
• Feedback: otcdmin@outlook.com, Instagram @ottcouture.eu, Discord discord.gg/GMMSqePfPh.
• License: AGPL-3.0. Private use and developer testing are allowed; CSCs and companies must license commercial usage directly with ottcouture.eu.