v0.4.0 — Media transcoding · Postgres backend

What's new in v0.4.0

🎨 Media transcoding (Pillow-backed)

Per-platform image resize for 10 platform specs out of the box: `instagram` (square), `instagram_portrait` (4:5), `instagram_reels` (9:16), `twitter` (16:9), `linkedin` (1200x627), `bluesky`, `telegram`, `threads`, `youtube_thumbnail`, `youtube_short_thumbnail`.

• Honors EXIF orientation (phone photos no longer come out sideways)
• Flattens RGBA/LA/P modes to RGB before JPEG encoding
• Centered aspect crop, no distortion
• Never upscales — quality trumps filling the spec
• LANCZOS resampling, progressive JPEG, optimize=True

REST:
```bash
curl -X POST "http://localhost:8000/media/transcode?platform=twitter" \
-H "Content-Type: image/jpeg" --data-binary @photo.jpg --output photo.twitter.jpg

curl http://localhost:8000/media/specs | jq .twitter
```

Python:
```python
from media import transcode_image
transcode_image("photo.jpg", "instagram") # → photo.instagram.jpg
```

🐘 Postgres queue backend

Opt in via `DATABASE_URL` for true ACID + cross-region multi-worker setups.

• Auto-creates schema on first use (table + partial index on queued rows)
• `mark_publishing` uses conditional UPDATE (`WHERE id = %s AND status = 'queued'`) — two workers in different regions can't double-claim
• `mark_failed` uses `attempts = attempts + 1` directly (no read-modify-write race)
• Factory falls back gracefully: `DATABASE_URL` → Postgres, else `REDIS_URL` → Redis, else JSONL
• `pip install open-dispatch[postgres]` for the optional psycopg2 dep

📜 OSS scaffolding

• `LICENSE` — MIT (matches what we'd advertised since day one)
• `CONTRIBUTING.md` — quick setup, project layout map, step-by-step "add a platform adapter" recipe, code style, commit conventions

Stats

• 129 tests passing (up from 102 in v0.3.0)
• 7 platforms supported, 10 image specs
• 3 queue backends (JSONL / Redis / Postgres) sharing one `QueueProtocol`
• 0 net runtime deps added beyond Pillow (Postgres + Redis remain optional extras)

Coming next

• TikTok adapter (still waiting on Content Posting API approval)
• Video transcoding (ffmpeg-backed, future)
• npm publish for the n8n node

v0.3.0 — YouTube Shorts · Redis backend · 102 tests

What's new in v0.3.0

📹 YouTube Shorts adapter

`adapters/youtube.py` — full Data API v3 resumable upload flow:

• OAuth2 refresh token flow (run once, store, never re-auth)
• Auto-appends `#Shorts` to description for correct YouTube classification
• Account-scoped refresh tokens
• Privacy whitelist (public/unlisted/private)
• Format key: `youtube_short` { video_path, title?, description?, tags?, privacy?, caption? }

The AI caption adapter also understands YouTube — emits `{title, description}` shape with 100/5000 char enforcement.

🔴 Redis queue backend

`api/queue.py` — opt-in via `REDIS_URL`:

• Multi-worker safe via guarded `queued → publishing` transition
• Fast `list_due` via ZSET keyed by scheduled_for epoch
• Single `QueueProtocol` interface across both backends — worker code unchanged

```bash
docker compose --profile redis up -d # bundles redis:7-alpine
```

`pip install open-dispatch[redis]` to add the optional dep.

✅ Adapter test coverage

30+ new unit tests in `tests/test_adapters.py` + `tests/test_redis_queue.py` — mock-based, no network, no real creds. Every platform exercised across missing-creds / empty-payload / happy-path / HTTP-error paths.

Stats

• 102 tests passing (up from 60 in v0.2.0)
• 7 platforms supported: Twitter / X, Bluesky, Telegram, Instagram, LinkedIn, Threads, YouTube Shorts
• 49 source files, 0 new runtime dependencies (Redis is optional)

Coming next

• TikTok adapter (still waiting on Content Posting API approval)
• Postgres queue (for cross-region multi-worker)
• Media transcoding (per-platform image/video resize)

v0.2.0 — Web UI · Threads · AI captions · n8n node

What's new in v0.2.0

Four major roadmap items shipped:

🖥️ Web UI

Dark terminal-themed dashboard powered by HTMX + Jinja2 — no JS build step, vendored htmx.min.js (51KB).

• GET / — queue dashboard with status filters and 5s auto-refresh
• GET /compose — post composer with platform checkboxes (configured-creds indicator), schedule, webhook, and ✦ Adapt with AI preview button
• GET /queue/{id} — row detail with retry, ContentUnit JSON viewer

The JSON API stays back-compat via content negotiation: `Accept: application/json` → dict, `Accept: text/html` → HTML page.

🧵 Threads adapter

Meta Threads Graph API integration — text, image, video posts. Account-scoped creds. Configurable settle delay for media (Meta recommends ~30s between container creation and publish).

✨ AI caption adapter

One source caption → 6 platform-native posts, respecting char limits and style conventions.

```bash
curl -X POST http://localhost:8000/ai/adapt -H "Content-Type: application/json" -d '{
"text": "Open-Dispatch v0.2 ships: web UI, Threads, AI caption rewriter.",
"platforms": ["twitter", "bluesky", "linkedin", "instagram", "threads"]
}'
```

Provider priority: Ollama (local) → OpenRouter (cloud) → heuristic (no LLM). The heuristic fallback means the endpoint never 500s on missing creds.

🔌 n8n community node

`n8n-nodes-open-dispatch` — single node with 5 operations: Dispatch, Adapt Caption with AI, Get Queue Row, Retry Queue Row, List Queue.

```bash
cd n8n-node && npm install --ignore-scripts && npm run build
```

Stats

• 60 tests passing (web UI: 9, AI adapter: 24, schema/queue/api: 25, Threads: 2)
• 6 platforms supported: Twitter / X, Bluesky, Telegram, Instagram, LinkedIn, Threads
• 0 production dependencies added beyond the minimum

Coming next

• TikTok adapter (waiting on Content Posting API approval)
• YouTube Shorts adapter
• Redis + RQ queue backend (multi-worker production setups)
• Media transcoding (per-platform image/video resize)