PRICING
Springy is free.
Your LLM bill is not.
MIT-licensed, no SaaS tier, no pricing page sleight-of-hand. You pay whatever your upstream providers charge — and nothing else. Here's exactly what that looks like in real dollars.
One anchor a week, fan-out to 5-7 platforms, no images, no video voiceover. Realistic for most solo creators.
Daily cascade, all 14 platforms, voice-RAG active, occasional hero image, ChatGPT Plus sub covers the Claude/GPT line.
3-5 brands running daily, images on, video voiceover via ElevenLabs, heavier critique loop (fine-grained critic enabled).
What adds up to that number.
| Upstream | What it powers | Required? | Typical monthly cost |
|---|---|---|---|
| Anthropic (Claude) | Generator + critic + humanizer + fine-grained critic. | One of these | $5–$40 |
| OpenAI (GPT-5 / Codex) | Same, swap via SPRINGY_LLM_PROVIDER=openai. | One of these | $0–$40 |
| ChatGPT Plus / Pro | OAuth via Codex sign-in — subscription covers inference. | If you already have it | $0 extra (your $20/mo sub) |
| Voyage (voyage-3) | Brand-voice RAG embeddings. | Strongly recommended | < $1 |
| Replicate (Flux) | Hero images per platform. | Optional | $0–$15 |
| ElevenLabs | TTS previews + video voiceover. | Optional | $0–$22 |
| OpenAI Whisper | Voice memo → anchor essay. | Optional | $0–$5 |
Hard caps are built in.
Per-platform ceiling on the critique loop. When total LLM spend for one platform's critique + revise attempts crosses this, the loop aborts and returns the current draft. Spend is tracked in the DB and returned even if the DB write fails — fail-closed, not fail-open.
Shows the last 30 days of spend broken down by provider × model × caller. Pass --trace <id> to see exactly what one cascade cost. Pass --daily for a day-by-day breakdown.
Knobs that move the number.
Use your ChatGPT subscription. If you're a Plus/Pro subscriber, springy auth openai → "Sign in with ChatGPT" runs every cascade on subscription credits. Zero marginal API cost. Most solo setups will land under $5/month total.
Drop the fine-grained critic for short-form. It doubles token spend on the anchor essay. Worth it for long-form where line-level feedback matters; overkill for a LinkedIn post.
Skip images until you need them. Visual hero images matter more for LinkedIn and Medium. Your cascade runs fine without them and saves ~$0.25 per run.
Use the cheap-tier classifier. The engagement loop (comment classification) uses Haiku / GPT-5-mini by default — pennies. If you widen to GPT-5 or Sonnet there, the bill can double without a quality gain.
Trim the fan-out. Set CASCADE_PLATFORMS=linkedin,x-thread,bluesky to only generate for 3 platforms. Cost scales roughly linearly with platform count.
Not paying for.
- ✓A SaaS subscription. Springy is MIT-licensed and runs on your machine.
- ✓A scheduling / posting intermediary like Buffer or Typefully.
- ✓A separate analytics tool. Local SQLite tracks posts, metrics, variants, spend.
- ✓A web host. Springy writes drafts + publishes; it doesn't need a server.
- ✓Per-seat fees. Run as many brands as you have local memory for.