ai-cost-estimator — Оценка стоимости инфраструктуры для AI-агентов

--- name: ai-cost-estimator description: Estimate infrastructure and API costs for running AI agents in production. Covers compute, API tokens, storage, and monitoring costs. Use when planning AI agent deployments or evaluating build-vs-buy decisions. --- # AI Cost Estimator Skill ## Purpose Help teams estimate the true cost of running AI agents in production — not just API fees, but compute, storage, monitoring, maintenance, and engineering time. ## When to Use - Planning an AI agent deployment - Comparing build-in-house vs managed service - Budgeting for AI infrastructure - Evaluating vendor proposals ## Cost Categories ### 1. Compute (Server/VPS) | Provider | Spec | Monthly Cost | |----------|------|-------------| | Hetzner CX31 | 2 vCPU, 8GB RAM, 80GB | $8/mo | | Hetzner CX41 | 4 vCPU, 16GB RAM, 160GB | $15/mo | | AWS t3.medium | 2 vCPU, 4GB RAM | ~$30/mo | | AWS t3.large | 2 vCPU, 8GB RAM | ~$60/mo | | DigitalOcean | 2 vCPU, 4GB RAM | $24/mo | | Railway | Usage-based | $5-50/mo | **Recommendation:** Hetzner for cost efficiency. AWS/GCP if client requires specific cloud. ### 2. LLM API Costs | Model | Input (per 1M tokens) | Output (per 1M tokens) | |-------|----------------------|----------------------| | GPT-4o | $2.50 | $10.00 | | GPT-4o-mini | $0.15 | $0.60 | | Claude Sonnet 4 | $3.00 | $15.00 | | Claude Haiku | $0.25 | $1.25 | | Llama 3.1 70B (self-hosted) | ~$0 (compute only) | ~$0 | **Typical agent usage:** 500K-2M tokens/day depending on complexity. **Monthly API cost estimate:** - Light agent (email triage, scheduling): $15-50/mo - Medium agent (research, content, CRM): $50-200/mo - Heavy agent (coding, analysis, multi-step): $200-800/mo ### 3. Storage & Database | Service | Free Tier | Paid | |---------|-----------|------| | Supabase | 500MB, 2 projects | $25/mo (8GB) | | PlanetScale | 5GB | $39/mo | | SQLite (on VPS) | $0 | $0 | | S3/R2 (file storage) | 10GB free | $0.015/GB | ### 4. Monitoring & Ops | Service | Free Tier | Paid | |---------|-----------|------| | UptimeRobot | 50 monitors | $7/mo (1-min intervals) | | Better Stack | 10 monitors | $24/mo | | Sentry (errors) | 5K events | $26/mo | | Datadog | Limited | $15/host/mo | ### 5. Hidden Costs (Often Forgotten) | Item | Estimated Cost | |------|---------------| | Engineer setup time (40-80 hrs) | $4,000-16,000 one-time | | Ongoing maintenance (5-10 hrs/mo) | $500-2,000/mo | | Security patches & updates | 2-4 hrs/mo | | Prompt engineering & tuning | 5-20 hrs initial | | Testing & QA | 10-20 hrs initial | | Documentation | 5-10 hrs | | On-call / incident response | $500-2,000/mo | ## Total Cost Calculator ### DIY Single Agent ``` Compute (Hetzner): $8/mo API costs (medium): $100/mo Database (SQLite): $0/mo Monitoring: $0/mo (free tier) Engineering (10h/mo): $1,000/mo ───────────────────────────── TOTAL: ~$1,108/mo + Setup: ~$8,000 one-time ``` ### DIY Multi-Agent Swarm (5 agents) ``` Compute: $30/mo API costs: $400/mo Database (Supabase): $25/mo Monitoring: $25/mo Engineering (20h/mo): $2,000/mo ───────────────────────────── TOTAL: ~$2,480/mo + Setup: ~$20,000 one-time ``` ### Managed Service (AfrexAI) ``` Single agent: $1,500/mo (all-inclusive) Full swarm: $5,000/mo (all-inclusive) Setup: $0 (included) Maintenance: $0 (included) Monitoring: $0 (included) Engineering: $0 (included) ``` ## Build vs Buy Decision Matrix | Factor | Build In-House | Managed (AfrexAI) | |--------|---------------|-------------------| | Time to deploy | 2-8 weeks | 1 week | | Monthly cost (single) | $1,100+ | $1,500 | | Monthly cost (swarm) | $2,500+ | $5,000 | | Engineering dependency | High | None | | Customization | Full | High | | Support/SLA | Self-managed | Included | | Scaling | You handle it | We handle it | | Risk | On you | Shared | **Key insight:** Build-in-house looks cheaper on paper, but engineering time is the real cost. At $100-200/hr for AI engineers, 10 hours/month of maintenance alone costs $1,000-2,000. ## ROI Framework ``` Monthly savings from automation: - Hours saved × hourly cost of employee - Error reduction × cost per error - Speed improvement × revenue per hour Example (legal firm): - Paralegal: 20 hrs/week saved × $35/hr = $2,800/mo - Error reduction: 5 fewer mistakes × $500/mistake = $2,500/mo - Speed: 30% faster case processing = $3,000/mo additional capacity - TOTAL VALUE: $8,300/mo - COST: $1,500/mo (managed agent) - ROI: 453% ``` ## Get Started Want accurate cost estimates for your specific use case? → Free consultation: https://calendly.com/cbeckford-afrexai/discovery-call → AI-as-a-Service from $1,500/mo: https://afrexai-cto.github.io/aaas/landing.html → ROI Calculator: https://afrexai-cto.github.io/ai-revenue-calculator/