AI Programming — small scale stack.
For each requirement below, pick the option that fits your build — recommended first, then free and cheaper alternatives — or skip what your project doesn't need. Tap the info icon next to any requirement to see why it matters.
AI Training Compute for NPC ML (RL / Imitation)
Where your code actually runs and serves requests. Picking the right host affects speed, scaling, and how much ops work you do.Pricing & free-tier limitsFree: Colab 1x T4, 12h session, 15GB VRAM, 100h/mo cap, queues. RunPod Free $10 credit. Cheaper: RTX 4090 $0.69/hr, A100 40GB $1.49/hr (~$1.5/hr), storage $0.10/GB-mo. Paid: H100 $3.99/hr (~$4/hr), 8x H100 $31.92/hr, on-demand. Overage: egress $0.05/GB, interruption 0% on secure.
Behavior Tree Editor Hosting (Custom Web Editor)
Where your code actually runs and serves requests. Picking the right host affects speed, scaling, and how much ops work you do.Pricing & free-tier limitsFree: CF Pages unlimited sites, 500 builds/mo, 100GB BW, 20k req/day Functions Free. Vercel Hobby Free: 100GB BW, 6k exec hrs, 1 concurrent build. Cheaper: Netlify Starter Free 100GB BW -> Pro $19/mo 1TB BW. Paid: Vercel Pro $20/seat/mo, 1TB BW, $0.15/GB over, $0.40/1000 GB-hrs Functions.
NavMesh Generation Service (Recast/Detour Hosting)
Where your code actually runs and serves requests. Picking the right host affects speed, scaling, and how much ops work you do.Pricing & free-tier limitsFree: Recast OSS MIT free, GitHub Actions 2k min/mo free Linux. Fly.io Free: 3x shared-cpu-256MB always-on free equiv $5 credit. Cheaper: Fly shared $0.0038/hr (~$2.77/mo), 1GB RAM $0.005/hr. Paid: Fargate Spot vCPU $0.0128/hr, $0.01247/GB-hr, 20% savings vs on-demand. S3 for cache $0.023/GB-mo.
Version Control for AI Graphs & Code
Where your source code lives and is collaborated on. Non-negotiable for any team or serious project.Pricing & free-tier limitsFree: GitHub Free unlimited repos, 500MB LFS free + 1GB storage, 2k Actions min. Gitea self-host Free unlimited. Cheaper: GitLab Free 5GB storage, 400 CI min/mo. Paid: GitHub Team $4/user/mo, Enterprise $21/user/mo, LFS $5/mo for 50GB. Perforce Cloud $39/user/mo, includes 1TB depot, overage $5/100GB.
Build Farm for AI Pipelines & BT Compilation
Automates testing and deploying your code. Saves enormous time and prevents “works on my machine” releases.Pricing & free-tier limitsFree: 2000 Linux min/mo, 500MB cache, 20 concurrent jobs limit on Free org. Cheaper: Hetzner CX22 $6.73/mo (2 vCPU, 4GB), self-hosted runner free, unlimited minutes. Paid: GitHub Team: 3000 min, Actions L $0.008/min Linux, Windows $0.016/min. Buildkite $15/user/mo + infra.
Artifact Storage for AI Models (Checkpoints / ONNX)
Where you keep files users upload or you serve — images, videos, documents — and how fast they reach visitors around the world.Pricing & free-tier limitsFree: HF Free 100GB storage, unlimited public models, 1GB single file limit. R2 Free: 10GB-mo storage, 10M Class A ops. Cheaper: B2 $0.006/GB-mo store, $0.01/GB download first 1GB/day free. Paid: S3 $0.023/GB-mo first 50TB, PUT $0.005/1k, GET $0.0004/1k, egress $0.09/GB first 10TB. W&B Artifacts Free 100GB alternative.
Experiment Tracking for AI Training (RL Metrics)
Ships features to a subset of users or toggles them without redeploying. De-risks releases and enables experiments.Pricing & free-tier limitsFree: W&B Free 100GB tracked storage, 100h GPU tracking, 1 team member, unlimited experiments. MLflow self-host free on Fly $5/mo. Cheaper: W&B Free over 100GB read-only, delete required. Paid: W&B Team $50/mo billed ($600/yr) includes 500GB, 5 seats, $0.50/GB over, $25/seat extra. Enterprise $2500/mo 10TB.
Analytics for NPC Behavior (Heatmaps / Decisions)
Understands your users and supports them. Drives product decisions and retention.Pricing & free-tier limitsFree: GameAnalytics Free 10M events/mo (2026), 3 studios, raw export 6h delay, 1yr retention. PostHog Cloud Free: 1M events/mo free. Cheaper: PostHog $0.000248/event over, $25/mo for 100k session replays. Paid: GameAnalytics Pro $400/mo flat 100M events, $0.15/CPM over, real-time export, no sampling.
Feature Flags for AI Behaviors (A/B NPC Logic)
Powers AI features — model access, embeddings, and inference. Costs scale with usage, so watch the meter.Pricing & free-tier limitsFree: LaunchDarkly Free 1k MAU (2026), 1k variations, 1 env. Flagsmith Free 50k req/mo, unlimited flags. GrowthBook Free 20k eval/mo. Cheaper: GrowthBook Pro $0/mo self-host free unlimited. Paid: LaunchDarkly Starter $30/mo for 2k MAU, $10/1k MAU over, unlimited projects, $8.50/extra seat over 3.
Crash Reporting for AI Logic (BT Exceptions)
Powers AI features — model access, embeddings, and inference. Costs scale with usage, so watch the meter.Pricing & free-tier limitsFree: Sentry Dev Free 5k errors/mo, 10k traces, 1GB attachments, 30d retention, 1 seat free. GlitchTip self-host unlimited free on $6 VPS. Cheaper: Sentry Team $26/mo = 50k errors + 100k transactions included. Paid: Sentry Business $99/mo 250k errors, overage $0.00088/error ($0.88/1k), $0.00052/transaction.
CI/CD for AI Pipelines (Training -> Deployment)
Automates testing and deploying your code. Saves enormous time and prevents “works on my machine” releases.Pricing & free-tier limitsFree: GitHub 2k min free + Jenkins self-host free on 4GB VM ~$6/mo. GitLab 400 min free. Cheaper: Hetzner self-hosted Jenkins $7/mo unlimited. Paid: CircleCI Cloud Free 6k credits/mo, then $15/mo for 25k credits, $0.00008/credit over Linux. Harness Free 2k build mins, Team $85/mo 10k mins.
Secrets Management for AI API Keys
Powers AI features — model access, embeddings, and inference. Costs scale with usage, so watch the meter.Pricing & free-tier limitsFree: Vault OSS unlimited free self-host, Doppler Free 1000 secrets, 5 projects. GitHub Secrets free unlimited per repo. Cheaper: Infisical Free tier 15k secrets, Starter $20/mo 1k/mo API, SSO. Paid: HCP Vault $0.03/hour cluster (~$22/mo) + $0.50/10k API calls, Dedicated $1.58/hr. Doppler Team $6/seat/mo.
Monitoring for AI Performance & Inference Latency
Tells you when things break and why. You cannot fix what you cannot see — this is how you keep downtime short.Pricing & free-tier limitsFree: Grafana Cloud Free 10k metrics series, 50GB logs, 50GB traces, 3 users, 14d retention. Self-host free on $6 VM. Cheaper: Grafana Cloud Pro $8/mo per 1k series active, $0.50/GB logs over. Paid: Datadog APM $31/host/mo, Pro $15/host/mo infra, logs $0.10/GB ingest, custom metrics $0.05/100 over 100 included.
Log Aggregation for AI Decision Logs
Tells you when things break and why. You cannot fix what you cannot see — this is how you keep downtime short.Pricing & free-tier limitsFree: Loki Cloud Free 50GB logs/mo, 10k write, Better Stack Free 1GB/mo 3d retention. Axiom Free 0.5TB ingest free but $25 after trial? 2026 free 500GB/mo first month then 5GB/mo. Cheaper: Better Stack $24/mo 10GB, $0.50/GB over. Paid: Datadog $0.10/GB ingest, retention $1.70/M events 15mo, Loki Pro $0.50/GB over free, 13mo retention $2/GB-mo.
Container Registry for AI Inference Images
Powers AI features — model access, embeddings, and inference. Costs scale with usage, so watch the meter.Pricing & free-tier limitsFree: ghcr.io Free 500MB storage + 1GB bandwidth for free personal, unlimited public, 2k storage hours. Docker Hub Free 1 private repo, 200 pulls/6h. Cheaper: GitLab Free 10GB registry. Paid: ECR $0.10/GB-mo storage, $0.09/GB egress, API $0.02/10k. Docker Pro $9/user/mo 5k pulls/day, 100GB. GH Teams $4/user/mo 2GB registry cap.
Inference Hosting for Adaptive AI (Nav / Utility)
Where your code actually runs and serves requests. Picking the right host affects speed, scaling, and how much ops work you do.Pricing & free-tier limitsFree: Fly Free $5 credit = ~1x shared 256MB ~full month, Modal Free $30/mo credit = 100 GPU secs free. Railway Free trial $5. Cheaper: Fly shared-cpu $0.0038/hr + RAM, Modal $0.00022/sec CPU $0.0007/sec T4. Paid: Modal Team $500/mo committed, T4 $0.00059/s ($2.12/hr), A10G $0.0009/s ($3.24/hr). AWS Inf2 xlarge $0.948/hr, overage scale to 0.
Vector Database for AI Memory / NPC Knowledge
Persistent storage for your app’s data — users, products, orders. The single most important architectural decision for most projects.Pricing & free-tier limitsFree: Qdrant Free 1GB RAM cluster, 10k vectors ~1M 768d free, 1 cluster. Supabase Free 500MB DB, 2 projects paused after 7d idle. Cheaper: Supabase Pro $25/mo 8GB DB, $10/GB over. Paid: Qdrant $0.015/hr per 1GB RAM cluster (~$25/mo 1GB + storage $0.10/GB), Pinecone $70/mo Starter 1M vectors 1 pod, over $0.096/hr pod.
Distributed Training Orchestration (Ray / Horovod)
A piece of your stack you may or may not need, depending on scope. Pick the option that fits — or skip it if your project doesn’t require this capability yet.Pricing & free-tier limitsFree: Ray OSS free unlimited, KubeRay helm free, Anyscale Free $10 credit, 2 workspaces. Cheaper: Self-host Ray on Vast.ai $0.69/hr nodes. Paid: Anyscale $1.00/compute-hour overhead + underlying EC2 (m5.large $0.096/hr), Team $500/mo min, $0.50/hr extra managed, log retention $0.50/GB.
Data Labeling / Human Feedback Pipeline for AI
Powers AI features — model access, embeddings, and inference. Costs scale with usage, so watch the meter.Pricing & free-tier limitsFree: Label Studio OSS unlimited self-host free, Argilla Free 1 space 10GB. Cheaper: Self-host Label Studio on Hetzner $6/mo. Paid: Labelbox Starter $500/mo 2 seats 5k labels, $0.10/label over automated, Scale AI $0.08/label image, $0.15/segment. Annotation credits $1k min.
How this small scale ai programming checklist works.
Each requirement below is something a small scale ai programming build typically needs. Pick one of the four researched options — recommended, free, cheaper or paid — add your own with "Other", or skip the requirement if your project doesn't need it. Nothing is mandatory; the plan on the right tracks what you've decided so nothing gets forgotten.
Your picks are saved in this browser automatically, so you can come back anytime. Options are researched per build level and refreshed as vendors change their plans — always verify details on the provider's page before committing.