Infrastructure Stack Reference

persona:

Static / Frontend

5 entries

Astro

USED

Ship content-heavy sites with near-zero JS while still using components.

need it when: Blog, docs, marketing site, or anything where SEO and load speed matter more than app-like interactivity.
skip it: You're building a dashboard or highly interactive app — reach for a SPA framework instead.

Open SourceFree

48k

Cloudflare Pages / Workers

USED

Deploy static and edge-compute sites globally without managing servers.

need it when: You want free-tier-friendly global deploys with edge functions and fast cold starts.
skip it: You need long-running processes or heavy GPU workloads — edge runtime won't fit.

ManagedFree

Tailwind CSS

Style consistently without writing ad-hoc CSS or maintaining a design system from scratch.

need it when: Any project where you want utility-first styling and small production CSS.
skip it: Design-heavy marketing work with a dedicated designer handing off full CSS.

Open SourceFree

85k

htmx

Add dynamic partial updates to server-rendered pages without a JS framework.

need it when: Server-driven apps that need interactivity but you refuse a node build pipeline.
skip it: Complex client state, offline support, or heavy animation.

Open SourceFree

43k

Alpine.js

Sprinkle client-side behavior (dropdowns, toggles, tabs) onto server-rendered HTML.

need it when: Small interactive islands inside otherwise-static pages.
skip it: Anything with routing, shared state, or >200 lines of client logic.

Open SourceFree

28k

Tech	Need it when	Skip it	License
AstroUSED	Blog, docs, marketing site, or anything where SEO and load speed matter more than app-like interactivity.	You're building a dashboard or highly interactive app — reach for a SPA framework instead.	Open SourceFree
Cloudflare Pages / WorkersUSED	You want free-tier-friendly global deploys with edge functions and fast cold starts.	You need long-running processes or heavy GPU workloads — edge runtime won't fit.	ManagedFree
Tailwind CSS	Any project where you want utility-first styling and small production CSS.	Design-heavy marketing work with a dedicated designer handing off full CSS.	Open SourceFree
htmx	Server-driven apps that need interactivity but you refuse a node build pipeline.	Complex client state, offline support, or heavy animation.	Open SourceFree
Alpine.js	Small interactive islands inside otherwise-static pages.	Anything with routing, shared state, or >200 lines of client logic.	Open SourceFree

Data / Storage

6 entries

SQLite

USED

Persist structured data without operating a database server.

need it when: Single-node apps, prototypes, embedded analytics, or anything under ~10 concurrent writers. It's the default answer.
skip it: Multi-region writes or genuinely high concurrent write volume.

Open SourceFree

— (bundled, ~6M LOC C)

Cloudflare R2

USED

Store blobs and files without egress fees eating your margin.

need it when: Object storage for uploads, backups, model artifacts, or static assets — zero egress cost is the hook.
skip it: You're all-in on AWS and S3 egress doesn't matter at your scale.

ManagedFree

Turso / libSQL

Distributed SQLite with replicas at the edge.

need it when: You want SQLite ergonomics but need read replication or a hosted story.
skip it: Simple single-server apps — plain SQLite is simpler.

Open SourceManaged

10k

PostgreSQL

The default relational database when you need real concurrency, extensions, and a long future.

need it when: Multi-tenant SaaS, complex queries, JSONB workloads, or anything you expect to live 10 years.
skip it: Single-node tools where SQLite suffices — don't run a server you don't need.

Open SourceFree

17k

Redis / Valkey

Cache hot data, rate-limit, and coordinate workers with sub-millisecond latency.

need it when: Session storage, caching layers, queues under load, leaderboards.
skip it: You'd be adding a network hop to save 5ms — SQLite in WAL mode may be enough.

Open SourceFree

68k

DuckDB

Run analytical SQL directly over Parquet/CSV files without a warehouse.

need it when: Local analytics, log analysis, embedding dumps, ad-hoc data exploration.
skip it: OLTP workloads — it's columnar and built for scans, not transactions.

Open SourceFree

28k

Tech	Need it when	Skip it	License
SQLiteUSED	Single-node apps, prototypes, embedded analytics, or anything under ~10 concurrent writers. It's the default answer.	Multi-region writes or genuinely high concurrent write volume.	Open SourceFree
Cloudflare R2USED	Object storage for uploads, backups, model artifacts, or static assets — zero egress cost is the hook.	You're all-in on AWS and S3 egress doesn't matter at your scale.	ManagedFree
Turso / libSQL	You want SQLite ergonomics but need read replication or a hosted story.	Simple single-server apps — plain SQLite is simpler.	Open SourceManaged
PostgreSQL	Multi-tenant SaaS, complex queries, JSONB workloads, or anything you expect to live 10 years.	Single-node tools where SQLite suffices — don't run a server you don't need.	Open SourceFree
Redis / Valkey	Session storage, caching layers, queues under load, leaderboards.	You'd be adding a network hop to save 5ms — SQLite in WAL mode may be enough.	Open SourceFree
DuckDB	Local analytics, log analysis, embedding dumps, ad-hoc data exploration.	OLTP workloads — it's columnar and built for scans, not transactions.	Open SourceFree

Auth

5 entries

Clerk

Ship complete auth (sessions, MFA, orgs, SSO) without becoming an auth engineer.

need it when: SaaS products where auth is table stakes and you'd rather build the product. Orgs/multi-tenancy out of the box.
skip it: Deep customization needs, strict data-residency rules, or you want zero vendor lock on identity.

ManagedFree

Auth.js (NextAuth)

OAuth/session auth that lives in your app, not a vendor's cloud.

need it when: JS/TS apps where you want provider logins with full control and no external dependency.
skip it: Non-JS backends or when you need enterprise SSO on day one.

Open SourceFree

26k

Keycloak

Self-hosted identity with SAML, OIDC, and enterprise federation.

need it when: Compliance-driven environments, on-prem requirements, or B2B SSO at scale.
skip it: Solo projects and MVPs — its operational weight will bury you.

Open SourceSelf-Host

27k

WorkOS

Add enterprise SSO/SCIM to an existing app without rebuilding auth.

need it when: You're selling to companies whose IT departments demand SAML and directory sync.
skip it: Pre-enterprise-stage products — it's a later-layer purchase.

Managed

Lucia (pattern)

Roll minimal session auth with plain primitives — no framework magic.

need it when: You understand auth and want ~200 lines of readable session code you fully own.
skip it: You want OAuth providers handled for you — use Auth.js instead.

Open SourceFree

10k

Tech	Need it when	Skip it	License
Clerk	SaaS products where auth is table stakes and you'd rather build the product. Orgs/multi-tenancy out of the box.	Deep customization needs, strict data-residency rules, or you want zero vendor lock on identity.	ManagedFree
Auth.js (NextAuth)	JS/TS apps where you want provider logins with full control and no external dependency.	Non-JS backends or when you need enterprise SSO on day one.	Open SourceFree
Keycloak	Compliance-driven environments, on-prem requirements, or B2B SSO at scale.	Solo projects and MVPs — its operational weight will bury you.	Open SourceSelf-Host
WorkOS	You're selling to companies whose IT departments demand SAML and directory sync.	Pre-enterprise-stage products — it's a later-layer purchase.	Managed
Lucia (pattern)	You understand auth and want ~200 lines of readable session code you fully own.	You want OAuth providers handled for you — use Auth.js instead.	Open SourceFree

Backend / API

6 entries

Hono

USED

Ultra-light, fast HTTP framework that runs on any JS runtime (Workers, Node, Bun, Deno).

need it when: APIs on Cloudflare Workers or anywhere cold starts matter. The default choice for this stack.
skip it: You need a batteries-included framework with ORM, auth, and admin built in.

Open SourceFree

23k

FastAPI

Typed Python APIs with automatic OpenAPI docs and first-class async.

need it when: ML/AI services, data pipelines exposed as APIs, or any Python-heavy backend.
skip it: Pure JS shops — Hono covers the same ground.

Open SourceFree

78k

Drizzle ORM

Type-safe SQL query building without a heavy runtime or query-engine binary.

need it when: TypeScript apps where you want SQL-shaped APIs and migrations that don't fight you.
skip it: Rapid prototyping where Rails-style magic outweighs control.

Open SourceFree

30k

tRPC

End-to-end typed APIs without writing a schema layer or codegen step.

need it when: Full-stack TypeScript monorepos where client and server share types naturally.
skip it: APIs consumed by non-TS clients or third parties — use REST/OpenAPI.

Open SourceFree

36k

Go (net/http / stdlib)

Single-binary backends with predictable performance and minimal dependencies.

need it when: CLI-adjacent services, high-throughput proxies, or anything you deploy as one static binary.
skip it: You live in Python/TS land and velocity matters more than binary size.

Open SourceFree

Railway / Fly.io

Deploy containers and full backends without building a platform team.

need it when: You've outgrown serverless or need long-running processes, queues, and persistent volumes.
skip it: Static + edge is enough — Cloudflare covers it cheaper.

ManagedFree

Tech	Need it when	Skip it	License
HonoUSED	APIs on Cloudflare Workers or anywhere cold starts matter. The default choice for this stack.	You need a batteries-included framework with ORM, auth, and admin built in.	Open SourceFree
FastAPI	ML/AI services, data pipelines exposed as APIs, or any Python-heavy backend.	Pure JS shops — Hono covers the same ground.	Open SourceFree
Drizzle ORM	TypeScript apps where you want SQL-shaped APIs and migrations that don't fight you.	Rapid prototyping where Rails-style magic outweighs control.	Open SourceFree
tRPC	Full-stack TypeScript monorepos where client and server share types naturally.	APIs consumed by non-TS clients or third parties — use REST/OpenAPI.	Open SourceFree
Go (net/http / stdlib)	CLI-adjacent services, high-throughput proxies, or anything you deploy as one static binary.	You live in Python/TS land and velocity matters more than binary size.	Open SourceFree
Railway / Fly.io	You've outgrown serverless or need long-running processes, queues, and persistent volumes.	Static + edge is enough — Cloudflare covers it cheaper.	ManagedFree

Search / CMS

5 entries

Pagefind

USED

Full-text search for static sites, generated at build time, zero server.

need it when: Docs sites and blogs with hundreds to thousands of pages. Ships a chunked index loaded on demand.
skip it: Dynamic content or user-generated data — you need a live index.

Open SourceFree

14k

Ghost

USED

Publishing platform with memberships, newsletters, and a real editor out of the box.

need it when: A blog/publication where writing workflow matters more than building one from markdown files.
skip it: Docs or product content that belongs in your repo and version control.

Open SourceManagedSelf-Host

48k

Meilisearch

Fast, typo-tolerant search you can self-host in minutes.

need it when: Product search, docs search, or any dataset under a few million docs with instant results.
skip it: Complex relevance engineering or huge corpora — that's Elasticsearch/Tantivy territory.

Open SourceSelf-HostManaged

49k

Tantivy

Full-text search engine library in Rust — embed search in your own binary.

need it when: You want Lucene-class search without JVM ops, embedded in your service.
skip it: You want a turnkey server — use Meilisearch.

Open SourceFree

13k

MDX / Content Collections

Keep content in the repo, type-checked and version-controlled.

need it when: Docs, changelogs, and technical writing that belongs next to the code.
skip it: Non-technical authors need a WYSIWYG editor — use Ghost or a headless CMS.

Open SourceFree

Tech	Need it when	Skip it	License
PagefindUSED	Docs sites and blogs with hundreds to thousands of pages. Ships a chunked index loaded on demand.	Dynamic content or user-generated data — you need a live index.	Open SourceFree
GhostUSED	A blog/publication where writing workflow matters more than building one from markdown files.	Docs or product content that belongs in your repo and version control.	Open SourceManagedSelf-Host
Meilisearch	Product search, docs search, or any dataset under a few million docs with instant results.	Complex relevance engineering or huge corpora — that's Elasticsearch/Tantivy territory.	Open SourceSelf-HostManaged
Tantivy	You want Lucene-class search without JVM ops, embedded in your service.	You want a turnkey server — use Meilisearch.	Open SourceFree
MDX / Content Collections	Docs, changelogs, and technical writing that belongs next to the code.	Non-technical authors need a WYSIWYG editor — use Ghost or a headless CMS.	Open SourceFree

Observability

5 entries

GoatCounter

USED

Privacy-friendly page analytics without cookie banners or surveillance pricing.

need it when: Blogs and small sites where you want honest traffic counts and nothing creepy.
skip it: Funnel/behavior analytics for a product — you need event-based tooling.

Open SourceFreeSelf-Host

Grafana + Prometheus

Metrics collection and dashboards that are the industry default.

need it when: Any server infrastructure you operate yourself. Start here before buying anything.
skip it: Serverless-only stacks — you'll get more from provider-native logs/metrics.

Open SourceSelf-HostFree

67k / 57k

OpenTelemetry

Vendor-neutral tracing/metrics/logs instrumentation you won't have to rip out later.

need it when: Distributed systems where you need cross-service traces. Instrument once, export anywhere.
skip it: Single-service apps with log-based debugging — don't over-instrument.

Open SourceFree

— (multi-repo)

Sentry

Error tracking with stack traces, release correlation, and alerting.

need it when: Any production app with real users. The first hour of setup pays for itself forever.
skip it: Honestly, almost never — the free tier covers most small projects.

Open SourceManagedSelf-Host

40k

Axiom

Serverless-native log analytics with generous free tier and fast full-text queries.

need it when: High-volume logs from edge/serverless functions where traditional log tools cost too much.
skip it: You already run Grafana/Loki and it's working.

ManagedFree

Tech	Need it when	Skip it	License
GoatCounterUSED	Blogs and small sites where you want honest traffic counts and nothing creepy.	Funnel/behavior analytics for a product — you need event-based tooling.	Open SourceFreeSelf-Host
Grafana + Prometheus	Any server infrastructure you operate yourself. Start here before buying anything.	Serverless-only stacks — you'll get more from provider-native logs/metrics.	Open SourceSelf-HostFree
OpenTelemetry	Distributed systems where you need cross-service traces. Instrument once, export anywhere.	Single-service apps with log-based debugging — don't over-instrument.	Open SourceFree
Sentry	Any production app with real users. The first hour of setup pays for itself forever.	Honestly, almost never — the free tier covers most small projects.	Open SourceManagedSelf-Host
Axiom	High-volume logs from edge/serverless functions where traditional log tools cost too much.	You already run Grafana/Loki and it's working.	ManagedFree

AI / Agent Frameworks

6 entries

LangGraph

Build stateful, graph-structured agent workflows with checkpointing and human-in-the-loop.

need it when: Multi-step agents with branching logic, retries, and durable state you need to inspect.
skip it: Single-prompt LLM calls — a framework here is pure overhead.

Open SourceFree

13k

Pydantic AI

Type-safe agent framework built on Pydantic — structured outputs without string parsing.

need it when: Python teams that want agent logic with real validation and dependency injection.
skip it: JS/TS shops or graph-heavy workflows — LangGraph fits better.

Open SourceFree

10k

Vercel AI SDK

Streaming LLM UI primitives and provider abstraction for TypeScript apps.

need it when: Chat/streaming interfaces in React/Next with multi-provider model switching.
skip it: Backend-only pipelines with no streaming UI.

Open SourceFree

14k

Instructor

Extract structured data from LLM outputs with retries and validation.

need it when: Any pipeline where the LLM must return valid, schema-conformant JSON.
skip it: You've moved to a framework with native structured outputs.

Open SourceFree

10k

LiteLLM

One OpenAI-shaped interface over 100+ providers, with proxying, budgets, and fallbacks.

need it when: Multi-provider routing, cost controls, or dropping in new models without code changes.
skip it: Single-provider apps — call the SDK directly.

Open SourceFree

—

Hermes Agent

USED

First-party agent runtime for this stack: tool-calling loop with durable state on SQLite.

need it when: You want a minimal, inspectable agent loop you fully own — no framework lock-in, contract-testable.
skip it: You need a large ecosystem of prebuilt integrations — use LangGraph.

Open SourceSelf-Host

Tech	Need it when	Skip it	License
LangGraph	Multi-step agents with branching logic, retries, and durable state you need to inspect.	Single-prompt LLM calls — a framework here is pure overhead.	Open SourceFree
Pydantic AI	Python teams that want agent logic with real validation and dependency injection.	JS/TS shops or graph-heavy workflows — LangGraph fits better.	Open SourceFree
Vercel AI SDK	Chat/streaming interfaces in React/Next with multi-provider model switching.	Backend-only pipelines with no streaming UI.	Open SourceFree
Instructor	Any pipeline where the LLM must return valid, schema-conformant JSON.	You've moved to a framework with native structured outputs.	Open SourceFree
LiteLLM	Multi-provider routing, cost controls, or dropping in new models without code changes.	Single-provider apps — call the SDK directly.	Open SourceFree
Hermes AgentUSED	You want a minimal, inspectable agent loop you fully own — no framework lock-in, contract-testable.	You need a large ecosystem of prebuilt integrations — use LangGraph.	Open SourceSelf-Host

Agent Infrastructure / Orchestration

8 entries

MCP (Model Context Protocol)

USED

Standard protocol connecting models to tools, resources, and data sources.

need it when: You're building agents that need to interoperate with a growing ecosystem of tool servers.
skip it: Single-app tool calling — plain function calling is simpler.

Open SourceFree

MCP Supply-Chain Security

Tool descriptions are attacker-controlled input — they can steer an agent, leak context, or trigger unintended actions.

need it when: You run agents that connect to third-party MCP servers. Pin the trust boundary: allow-list servers, validate descriptions, treat every tool call as an input-validation gate.
skip it: You only call first-party tools you wrote yourself behind a stable interface.

FreeSecurityPrompt-InjectionTrust

Temporal

Durable workflow execution: code that survives crashes, retries, and multi-day processes.

need it when: Long-running agent pipelines, billing flows, or anything where 'the process died mid-way' is unacceptable.
skip it: Short-lived request/response work — a queue is enough.

Open SourceSelf-HostManaged

13k

Firecracker / gVisor

Run untrusted agent-generated code in hardened microVMs or sandboxes.

need it when: Agents that execute code. Non-negotiable isolation layer before anything else.
skip it: Never for agent code execution — if you're skipping this, you're skipping safety.

Open SourceSelf-HostFree

27k / 16k

E2B

Managed sandboxes for running AI-generated code with SDK access.

need it when: You need agent code execution now and don't want to operate Firecracker yourself.
skip it: High-volume execution where per-second sandbox billing dominates costs — self-host.

ManagedFreeOpen Source

pgvector

Vector similarity search inside PostgreSQL — embeddings without new infrastructure.

need it when: RAG under ~10M vectors where you want one database for everything.
skip it: Billion-scale retrieval or heavy ANN tuning — dedicated vector DBs pull ahead.

Open SourceFree

14k

Qdrant

Purpose-built vector database with filtering, quantization, and distributed sharding.

need it when: Large-scale semantic search or hybrid filtering over embeddings at production volume.
skip it: Small RAG apps — pgvector in your existing Postgres is the right boring answer.

Open SourceSelf-HostManaged

23k

Celery / BullMQ

Background job queues for work that shouldn't block a request.

need it when: Email, embedding generation, webhooks, scheduled tasks — anything async.
skip it: Durable multi-step workflows — upgrade to Temporal before you outgrow queues painfully.

Open SourceFree

26k / 16k

Tech	Need it when	Skip it	License
MCP (Model Context Protocol)USED	You're building agents that need to interoperate with a growing ecosystem of tool servers.	Single-app tool calling — plain function calling is simpler.	Open SourceFree
MCP Supply-Chain Security	You run agents that connect to third-party MCP servers. Pin the trust boundary: allow-list servers, validate descriptions, treat every tool call as an input-validation gate.	You only call first-party tools you wrote yourself behind a stable interface.	FreeSecurityPrompt-InjectionTrust
Temporal	Long-running agent pipelines, billing flows, or anything where 'the process died mid-way' is unacceptable.	Short-lived request/response work — a queue is enough.	Open SourceSelf-HostManaged
Firecracker / gVisor	Agents that execute code. Non-negotiable isolation layer before anything else.	Never for agent code execution — if you're skipping this, you're skipping safety.	Open SourceSelf-HostFree
E2B	You need agent code execution now and don't want to operate Firecracker yourself.	High-volume execution where per-second sandbox billing dominates costs — self-host.	ManagedFreeOpen Source
pgvector	RAG under ~10M vectors where you want one database for everything.	Billion-scale retrieval or heavy ANN tuning — dedicated vector DBs pull ahead.	Open SourceFree
Qdrant	Large-scale semantic search or hybrid filtering over embeddings at production volume.	Small RAG apps — pgvector in your existing Postgres is the right boring answer.	Open SourceSelf-HostManaged
Celery / BullMQ	Email, embedding generation, webhooks, scheduled tasks — anything async.	Durable multi-step workflows — upgrade to Temporal before you outgrow queues painfully.	Open SourceFree

Expansion Candidates

reviewed September 2026

L3 · Auth — open gaps

Passkeys-first flow. Current entries are OAuth/session oriented; WebAuthn-native onboarding is untested in this stack. Candidate: a thin passkey layer over Lucia-style sessions.
Machine-to-machine auth. Agent-to-agent and agent-to-API authentication (mTLS, scoped tokens) has no first-class entry yet. This matters more as L8 grows.
Authorization (not authentication). Nothing here handles fine-grained permissions — Oso/OpenFGA-class tooling is the known gap.

L4 · Backend — expansion ideas

Queue-native runtime. Cloudflare Queues vs. self-hosted BullMQ needs a head-to-head under real agent workload shapes.
Rust sidecar layer. Hot-path components (embedding rerank, token counting) as Rust binaries called over UDS — evaluation pending.
API gateway. Rate limiting and key management for multi-tenant AI endpoints is currently ad hoc. Candidate: standalone gateway or Hono middleware consolidation.

Gaps are reviewed quarterly. Last pass: September 2026. Nothing was promoted this cycle — all candidates still in evaluation.