💬 Commit message: Update 2026-02-15 23:19:02, 1 files, 116 lines

📁 Files changed: 1
📝 Lines changed: 116

  • CLAUDE.md
This commit is contained in:
Adam Ladachowski
2026-02-15 23:19:02 +01:00
parent c9d76dc382
commit 023eae66d7
+106 -10
View File
@@ -4,7 +4,44 @@ This file provides guidance to Claude Code (claude.ai/code) when working with co
## Project Overview ## Project Overview
`tensors` is a Python CLI tool (`tsr`) for reading safetensor file metadata and interacting with the CivitAI API (search, fetch, download models). Built with Typer, Rich, and httpx. `tensors` is a Python CLI + FastAPI server for:
- Reading safetensor file metadata
- Searching CivitAI and HuggingFace models
- Managing a local model database (SQLModel ORM)
- Image gallery management
- Model downloads with resume support
Built with: Typer, Rich, FastAPI, httpx, SQLModel.
## Deployment
### Production (junkpile server)
| Service | Port | URL | Status |
|---------|------|-----|--------|
| tensors API | 51200 | https://tensors-api.saiden.dev | `systemctl status tensors` |
| ComfyUI | 8188 | internal only | `systemctl status comfyui` |
**Note:** ComfyUI runs separately - no integration with tensors yet.
```bash
# Deploy to junkpile
/deploy # or: ssh junkpile "cd /opt/tensors && git pull && pip install -e '.[server]' && systemctl restart tensors"
# Check status
curl https://tensors-api.saiden.dev/status
ssh junkpile "systemctl status tensors"
ssh junkpile "journalctl -u tensors -f" # logs
```
### TypeScript Client
Package: `@saiden/tensors` (GitHub: saiden-dev/tensors-typescript)
```bash
# Regenerate client after API changes
ruby ~/.claude/scripts/commands/tensors/generate_client.rb
```
## Commands ## Commands
@@ -21,29 +58,88 @@ just test # pytest with coverage
just fix # auto-fix lint + format just fix # auto-fix lint + format
just types # mypy only just types # mypy only
# Run server locally
uv run tsr serve --port 8000
# Run a single test # Run a single test
uv run pytest tests/test_tensors.py::TestClassName::test_name -v uv run pytest tests/test_tensors.py::TestClassName::test_name -v
``` ```
## Architecture ## Architecture
Five modules with clean separation: ### CLI (`tsr`)
- **`cli.py`** — Typer CLI with commands: `info`, `search`, `get`, `dl`, `config`. Legacy mode auto-converts bare `tsr file.safetensors` to `tsr info`. Entry point: `tsr = "tensors:main"` (pyproject.toml)
- **`api.py`** — CivitAI REST API wrapper using httpx. Search, fetch model/version/by-hash, download with resume support and Rich progress bars.
- **`config.py`** — XDG-compliant paths (`~/.config/tensors/config.toml`, `~/.local/share/tensors/`). Enums with `to_api()` methods for CivitAI parameter mapping. API key resolution: env var → config file → legacy `~/.sftrc`.
- **`safetensor.py`** — Binary safetensor header parsing (8-byte u64 LE header size → JSON metadata). SHA256 streaming hash computation.
- **`display.py`** — Rich table formatting for all output types. All major commands support `--json` output.
Entry point: `tsr = "tensors:main"` (pyproject.toml `[project.scripts]`). Commands: `info`, `search`, `get`, `dl`, `config`, `serve`
### Modules
| Module | Purpose |
|--------|---------|
| `cli.py` | Typer CLI commands |
| `api.py` | CivitAI REST API wrapper (httpx) |
| `hf.py` | HuggingFace Hub integration |
| `config.py` | XDG paths, enums, API key resolution |
| `safetensor.py` | Binary header parsing, SHA256 |
| `display.py` | Rich table formatting |
| `models.py` | SQLModel ORM (17 tables) |
| `db.py` | Database operations |
### Server (`/api/*`)
| Endpoint | Description |
|----------|-------------|
| `/api/search` | Unified CivitAI + HuggingFace search |
| `/api/civitai/model/{id}` | CivitAI model info |
| `/api/db/*` | Local database CRUD |
| `/api/images/*` | Gallery management |
| `/api/download/*` | Model downloads |
Auth: `X-API-Key` header (configured via `TENSORS_API_KEY` env var)
### Database
SQLite with SQLModel ORM. Tables:
- Local files: `local_files`, `safetensor_metadata`
- CivitAI cache: `models`, `model_versions`, `creators`, `tags`, `version_files`, `file_hashes`, `trained_words`, `version_images`, `image_generation_params`, `image_resources`
- HuggingFace cache: `hf_models`, `hf_model_tags`, `hf_safetensor_files`
## Code Standards ## Code Standards
- Python 3.12+, strict mypy, line length 130 - Python 3.12+, strict mypy, line length 130
- Ruff with extended rule set (E, W, F, I, B, C4, UP, ARG, SIM, TCH, PTH, PL, RUF) - Ruff with extended rule set (E, W, F, I, B, C4, UP, ARG, SIM, TCH, PTH, PL, RUF)
- PLR0913 (too many arguments) is intentionally ignored for CLI commands - PLR0913 (too many arguments) is intentionally ignored for CLI commands
- Tests use respx for HTTP mocking and a `temp_safetensor` fixture from conftest.py - Tests use respx for HTTP mocking and pytest fixtures from conftest.py
## Testing
```bash
# Full test suite
just test
# Specific test file
uv run pytest tests/test_db.py -v
# Test with coverage
uv run pytest --cov=tensors --cov-report=term-missing
# Test API endpoints locally
uv run tsr serve --port 8000 &
curl http://localhost:8000/status
curl -H "X-API-Key: test" http://localhost:8000/api/db/stats
```
## Release ## Release
Tags trigger PyPI publish: `git tag v0.1.x && git push origin v0.1.x`. See RELEASE.md for binary builds (Nuitka). Tags trigger PyPI publish: `git tag v0.1.x && git push origin v0.1.x`
## TODO: ComfyUI Integration
ComfyUI is running on junkpile (port 8188) but NOT integrated with tensors.
Needed:
- [ ] Add `/api/generate` endpoint that proxies to ComfyUI
- [ ] Workflow template management
- [ ] Queue job → poll for completion → return image
- [ ] Config: `[comfyui] url = "http://127.0.0.1:8188"`