gogogo

Landing page · Docs · Releases · CI

Your stack is ready. Come build. Every decision favors practical outcomes over abstract ideals. The stack optimizes for simplicity, consistency, and shipping software with minimal friction.

Every web project starts with the same conversation — pick a database, auth, router, reactive frontend, task queue… — and the project stalls at the decisions, installations, and configurations. Not at the code. This template answers those decisions once, ships them wired together, and documents how to change or remove each piece.

What it is not: a framework. Frameworks call your code and keep you on their release train; gogogo hands you wired, documented code and gets out of the way.

📖 Docs (single source of truth)

The manual lives on the site, not in this file: calionauta.github.io/gogogo/docs/

Page index (20 pages)

One-liners mirror the docs site manifest (site/build.mjs).

Get started

• overview — what gogogo is, who it is for, and the seven async layers it ships.
• motivation — the thesis: Go for almost everything, Zig as the named exception.
• getting-started — clone, run, first five minutes, and the commands you need.

Core

• architecture — directory layout, dependency direction, entry points, and route wiring gotchas.
• stack-layers — every dependency and why it is in the box.
• async-layers — the seven async layers, SSE Hub record-sync vs ephemeral signals, cross-instance and offline sync.
• features — every capability and its runtime opt-out.
• scope-taxonomy — Core / Plugin / Feature: the rule for deciding what is safe to delete.
• configuration — every environment variable and runtime constant.
• exception-to-go — Go-first policy: why Zig is an exceptional native boundary, plus the agent decision procedure.

Frontend

• todo-example — the Todo reference implementation and the contract to imitate.
• ui-skins — pluggable DaisyUI / Basecoat skins and the plugin contract.
• ui-sounds — vendored cuelume sound feedback and its accessibility contract.

Ship

• deploy — server layout, first-time setup, deploy workflow, build pipeline and version badge.
• desktop-mobile — Wails v3 desktop/mobile, edge sync, and the native window.
• admin-dashboard — admin surfaces, the single-cookie session, and the DagNats console.
• llm-and-credits — GoAI configuration and the optional ai-credits / BYOK plugin.

Operate

• local-ci — the five-tier feedback loop and make signoff.
• code-quality — the 33 linters, how to scope them, the Datastar-specific rules, and the ruleguard rules that encode project footguns as CI failures.
• troubleshooting — symptoms and where they actually come from.
• dagnats-bootstrap-workaround — the upstream DagNats bug behind the trigger console, and the removable seed that works around it.

Plus ARCHITECTURE.md — the canonical annotated dependency graph, and AGENTS.md — the working rules for human and AI agents in this repo. The rules an agent is meant to load live in skills/gogogo-coding-standards (Go + template standards, concurrency, perf, testing, the exception-to-go gate) — install it with npx skills add calionauta/gogogo.

---

🚀 Quick start

This repo is a GitHub template — click Use this template to get your own copy, then rename it. Do not build inside a clone of the template itself: your module path would stay github.com/calionauta/gogogo and deploys would target the template's server directory.

# 1. Scaffold with the installer: one command installs (binaries +
#    Go toolchain when missing), asks name/owner/plugins/features, renames,
#    trims what you skip, writes AGENTS.md, then `make dev` takes this
#    terminal — Ctrl-C stops dev. Needs curl + git, nothing else.
curl -sSfL https://raw.githubusercontent.com/calionauta/gogogo/master/install.sh | sh -s -- --run

# With Go already installed, the equivalent without curl|sh:
# go run github.com/calionauta/gogogo/cmd/gogogo@latest --run

# Manual fallback (no installer):
# gh repo create my-app --template calionauta/gogogo --clone
# cd my-app
# make rename NAME=my-app          # add OWNER=myorg if not under calionauta
# make site                        # regenerate the published docs
# make dev                         # run

This command needs Go and git (checked up front) — or curl + git via the binary install in CLI & MCP below, where install.sh bootstraps Go when missing. Open http://localhost:8080 for the landing page, then http://localhost:8080/todo for the demo (sign in with the seeded demo1@demo.app / demo1234456; demo2@demo.app takes the same password for two-user contention).

make rename only rewrites this project's identity. Sibling repos under the same owner (ai-credits, datastar-lint, pi-leakguard) are real dependencies and are left alone.

One make build compiles everything. No build tags, no feature matrix. Opt out at runtime with env vars (NATS_ENABLED=false, DAGNATS_ENABLED=false).

In the box

Layer	Choice	Why
Language	Go 1.27	Fast compilation, easy deploy, lean runtime. Go-first: Zig only as an exceptional native escape hatch (exception-to-go; the why: motivation)
Zero-config database and auth	Turso Database (SQLite-compatible, sqlc) + owned argon2id sessions	Zero-config Turso, cookie sessions, file-local storage — no separate service
Access control without a service	Casbin + users.role	Roles on the user row, policy in one model, enforced before the handler; embedded policy by default, DB-backed when evolving (rbac)
Reactive UI with no JS framework	Templ + Datastar (SSE) + Tailwind v4	Server-rendered HTML, ~12 KiB client, no JS framework build step
Two UI skins in one binary	DaisyUI v5 (default) + Basecoat	Both compiled into one binary, switchable at runtime
Background jobs with retry	goqite + SSE Hub	goqite executes with backoff; SSE Hub streams progress to the browser. No Redis
Workflows that survive a restart	DagNats	Multi-step workflows as declarative JSON; resume after a mid-run kill
Room ownership without races	GoAkt	One grain per room: heartbeat roster plus exactly-one presenter lock, supervised
Shared state that merges offline edits	Loro CRDT + Rough.js	Conflict-free state with no last-write-wins data loss
Realtime for one instance or many	SSE Hub + NATS JetStream	Per-user record push, plus cross-instance broadcast
AI features on any provider	GoAI + optional ai-credits	Any OpenAI-compatible provider; optional billing ledger + BYOK
Secrets encrypted at rest	age + ~/.secrets/	Local encryption. No vault, no cloud
Desktop apps on the same backend	Wails v3 + gogpu/ui	Native shells over the same backend, plus a no-HTTP frontend
Linting tuned for AI-written code	golangci-lint (33) + datastar-lint	Tuned for LLM-authored code: unchecked errors, context leaks, lost context propagation

Full table with the reasoning behind each choice: stack-layers. Client weight per feature (measured): 0–48 KB of JS — the backend is the source of truth, the browser holds intent and cache.

Everything async your app will need

Seven complementary layers, already inside, each with its own opt-out:

Background jobs that update the UI   → goqite + SSE Hub (always on)
Workflows that survive a restart     → DagNats as JSON (DAGNATS_ENABLED=false)
Collaboration without data loss      → Loro CRDT (remove internal/collab/)
Live record updates, scoped per user → SSE Hub record-sync (always on)
Ephemeral UI signals                 → SSE Hub via Datastar (always on)
Broadcast across instances           → JetStream (NATS_ENABLED=false)
Room ownership without races         → GoAkt grains (GOAKT_ENABLED=false)

Records flow through the SSE Hub record event — scoped per user by the owner column, free of charge. The ephemeral channel is reserved for ephemeral signals (toasts, client counts, workflow progress). Mixing those two up is the most common mistake in Datastar apps; async-layers explains why the split matters.

Capabilities

Every capability is always compiled. Each has a documented opt-out.

Capability	Opt-out
Todo app + live record-sync, stacked toasts, sound feedback	delete features/todo/
Queue + retry with exponential backoff, SSE lastRetry signal	— (core)
AI Suggest via GoAI, or keyless simulated LLM for the demo	no GOAI_API_KEY and SIMULATE_LLM=false
Ask AI: model renders catalog UI (cards, tables, sections) via SSE	trim genui (bundle with Chat), or no GOAI_API_KEY and SIMULATE_LLM=false (demo answers keyless)
Chat: conversation over the same catalog, persisted per owner	trim genui (bundle with Ask AI), or no GOAI_API_KEY and SIMULATE_LLM=false
Review loop: writer → critic → classifier judge, streamed verdict	trim review, or no GOAI_API_KEY (real-LLM only; page shows a setup hint)
AI credits + BYOK relay + Stripe top-ups	CREDITS_ENABLED=false
Collaborative whiteboard (CRDT + presence + offline outbox)	trim whiteboard (bundle with notes + collab transport), or COLLAB_SNAPSHOTS=memory
Room presence demo (GoAkt grains: roster + presenter lock)	delete features/room/
Shared notes (server-owned Loro Text, live peer text)	delete features/notes/
Durable WelcomeOnboarding workflow	DAGNATS_ENABLED=false
Hybrid offline sync (Service Worker + Leaf Node)	OFFLINE_SYNC_ENABLED=false
Single-store persistence (Turso Database via sqlc)	— (core)
Multi-instance realtime	NATS_ENABLED=false
Public landing page + auth-gated config view	delete features/landing/, features/config/
Wails v3 desktop/mobile + native window	delete cmd/desktop/, cmd/gui/

Infrastructure opts out with an env var. The installer automates the common deletions — named presets included: go run ./cmd/gogogo --preset minimal (core + todo + auth), no-ai, no-collab — see cmd/gogogo/README.md. Product features also opt out by deleting the package and its one wiring call — see features.

Core / Plugin / Feature

Every non-test .go file under internal/ and features/ carries a SCOPE annotation declaring its removal class. make check-scope enforces it in CI and in your pre-commit hook.

Annotation	Meaning	You would
SCOPE:core 🔴	The binary does not work without it	Customize, never remove
SCOPE:plugin 🟡	The binary works but loses a capability	Swap, or delete + remove the wiring call
SCOPE:feature 🟢	A demo or add-on	Keep as reference, then remove

The rule for agents: never delete a core file without asking; never keep a feature file in production if the domain doesn't need it. Full taxonomy and per-package removal table: scope-taxonomy.

For AI agents

Three entry points, one engine (internal/installer — CLI and MCP execute the same code path with the same receipts and proof):

• llms.txt — the agent entry index: what to read per task, invariants to not relearn. Published map of the docs site: site/llms.txt (index) + site/llms-full.txt (complete texts).
• Installer CLI — go run ./cmd/gogogo advise --need "..." for opinions (reads nothing, changes nothing), --preset minimal|no-ai|no-collab or --plugins/--features keep lists + --dry-run to preview, gogogo add <unit> to reverse a trim. Full reference: cmd/gogogo/README.md.
• MCP server — cmd/gogogo-mcp (separate module, stdio): advise_stack, capabilities_list, trim_plan, trim_apply (requires confirm:true), check_tree, add_unit. Tool table in cmd/gogogo-mcp/README.md.

Try it live

What	URL
Todo & Whiteboard demo	gogogo.calionauta.com
Durable workflow console (DagNats)	gogogo.calionauta.com/dagnats/

Sign in as demo1@demo.app / demo1234456 (demo2@demo.app takes the same password — open a second browser or incognito window to be both at once and prove contention, shared-vs-private todos, and authorship).

Commands

make dev           # Live reload with Air (regenerates templ + CSS)
make build         # Build binary (unified: everything included)
make ci-local      # Full local gate (= CI): templ, lint, checks, build, boot smoke, parallel fan-out (race + browser smokes)
make signoff       # ci-local + gh signoff stamp — the recommended pre-push gate
make setup         # Activate lefthook git hooks + verify agent tooling
make site          # Rebuild the docs site from docs/*.md

Full command reference: getting-started.

CLI & MCP (agent paths)

Humans get the guided installer above. Agents get the same engine (internal/installer) through two scripted paths — opinions first, preview second, mutation last. Full guides live in their own section: docs/cli.md and docs/mcp.md.

Path	Use when	Entry point
advise	deciding what to keep (use-case → keep/drop + off-switches + Go/Zig rule). Changes nothing	go run ./cmd/gogogo advise --need "..." --format json
trim --dry-run	previewing the scaffold as one JSON document	--no-tui --dry-run --format json (add --yes to apply)
add / --check	restoring a unit into a scaffolded checkout / verifying markers	add <unit> --from <pristine> --dir <proj> [--dry-run|--yes]
MCP server	tool calls instead of shell (advise_stack, capabilities_list, trim_plan, trim_apply, check_tree, add_unit)	cmd/gogogo-mcp (own module, stdio) + one JSON block in the client config

Destructive paths refuse without explicit confirmation (--yes / confirm:true). Strategy and the foreign-codebase boundary: getting-started §0b–0c.

No Go on this machine? Tagged releases ship static binaries (gogogo + gogogo-mcp, checksummed) — and install.sh bootstraps a user-space Go toolchain when none is found (~/.local/go, no sudo, existing installs untouched). It does not symlink go into ~/.local/bin: that directory is on your PATH, so the link would win over a system Go. It exports ~/.local/go/bin for the install and prints the line to persist it. Same one-liner as the quick start above (it is the quick start — append anything after --run and it reaches gogogo):

Before you push

make ci-local locally is the same gate CI runs, and make signoff adds a git stamp. It catches ~95% of regressions in under 3 minutes instead of waiting on a CI round trip. The tier ladder (T1 format/build → T5 signoff) and what each tier catches: local-ci.

Contributing

The most valuable contribution right now is using gogogo for a real project and reporting what broke — see CONTRIBUTING.md.

AI-assisted contributions are welcome, with a human in the loop: disclose tool use in the PR body (no AI trailers in commits or changelogs), explain motivation, include a reproduction, keep the diff minimal, and run make signoff before pushing. Please discuss features in an issue before opening a PR.

Acknowledgements

Inspired by northstar by Zangster — a Go + NATS + Datastar + Templ + DaisyUI application starter. See docs/inspirations.md for full credits and learnings.

License

MIT — see LICENSE. Open to feedback, PRs, and adaptations. If something doesn't make sense, if the stack doesn't fit your problem, or if you have a better idea, open an issue.