EvoLink — Unified Media + LLM Backup Skill

evolink.ai aggregator (153 models). Role in arsenal: backup layer for the media/LLM golden chain — invoked when a direct provider fails, or when a rare model is needed (Sora-2-Pro-Max, Kling o3, Seedance 1.5, Suno v5, nano-banana-pro).

• Paid credit (Dr. Wael charged it): ~656 credits remaining (live-checked). NOT free — treat as valuable.
• Base: https://api.evolink.ai/v1 · Auth: Authorization: Bearer $EVOLINK_API_KEY · OpenAI-compatible.
• Run with /usr/bin/python3. Source env.sh first OR the script self-parses env.sh.

Verified live (2026-06-17)

• credits ✅ · models ✅ (153) · chat ✅ (gpt-5.4 → "BACKUP_OK") · image ✅ (nano-banana → real URL, 1.6 cr, ~12s async).

Usage

bash
source /data/.openclaw/secrets/env.sh
python3 evolink.py credits
python3 evolink.py models --filter video
python3 evolink.py chat  --model gpt-5.4 --prompt "..." --max 500
python3 evolink.py image --model nano-banana-pro-beta --prompt "..."
python3 evolink.py video --model seedance-1.5-pro --prompt "..." [--image URL]
python3 evolink.py music --model suno-v5-beta --prompt "..."
python3 evolink.py status --task-id <id>

All output JSON: {"ok":true,...} / {"ok":false,"error":...,"status":N}.

Live pricing (verified)

Seedance 1.5 Pro $0.0247/s (cheapest) · Kling 3.0 $0.075/s · Sora-2 $0.08/s · Sora-2-Pro $0.639/10s · Veo 3.1 Lite $0.168/video.

Failover law (when to use)

Direct high-quality FIRST (Veo/Grok/GPT-image/HeyGen/golden LLM chain) → EvoLink when direct fails OR rare model needed → Fal/Replicate → cheap/local. Prefer cheapest-at-equal-quality. See DOMAIN_MODEL_RANKINGS.md § Aggregator Backup + NEW_PLATFORMS_2026-06-17.md.

Notes

• Media endpoints are async (return task id → poll). The tool polls automatically (image ~12s, video up to 7min).
• Reasoning models (gemini-3.5-flash) may need higher --max or they spend tokens on hidden reasoning and return empty.
• Credits guard: a weekly probe warns when balance is low (see master plan).