SunsetAI
§1 · AI‑NATIVE AGENCYMISSIONS OPEN
AI Engineering Workshop

Specs in. Shipped product out.
In weeks.

Abstract

We're an AI-native agency. We move this fast because we built sunset — our open framework — and we run every mission on it.

Request a mission→Open the framework
Inventors
Engineering staff, formerly of Meta; AssemblyAI; CrowdStrike; INRIA.
Field of art
Applied LLM systems — retrieval, evaluation, deployment.
Sheet 1 of 2 · FIG. 1Drawings
[10][12][14][16][18]FIG. 1 — MISSION TOPOLOGY.SPEC SIGNED AT HORIZON, MODEL CORE IGNITES, OUTPUT RAYS FAN TO THE USER, MONITORING REFLECTS BELOW THE EVAL THRESHOLD.

FIG. 1A · METHOD OF MANUFACTURE · SHEET 2 OF 2

ASSEMBLY 02UNFAIR ADVANTAGE— SHIPPED IN WEEKS

Most AI agencies sell time. We sell shipped. The difference is that we don't rebuild the rails on every mission — we forged them once, into a framework called sunset, and we open-sourced it.

Six steps. Eight decisions already pinned. AI writes the how; humans control the what.

DRAWING No. SUNSET‑1A    SHEET 2 OF 2    SCALE 1:1EXPLODED ALONG SHIPMENT AXIS    TOL. ±2% ACC.
AXIS — SHIPMENT[01]SPECsunset.yamlThe contract.Signed before code.[02]SCAFFOLDsunset newEight decisionsalready pinned.[03]WIREsunset.servicesAuth · LLM · RAG —pre-built, imported.[04]IMPLEMENTClaude in railsAI writes the how,against the spec.[05]PROVISIONsunset provisionOne terraform apply.Cloud Run + SQL + PS.[06]SHIPgit push mainPR preview.Merge to main deploys.
SPEC → SHIPPED
CYCLE
2–6 WEEKS
DECISIONS
8 PINNED
LOCK‑IN
NONE
LICENSE
MIT
REV. C · CREW: META / ASSEMBLYAI / CROWDSTRIKE / INRIASUNSETAI · ENGINEERING DISCLOSURE

“Humans want to control the what, not the how.”

— MANIFESTO.md · sunset framework
Request a missionOpen the framework

PRIOR ART: GITHUB SPEC‑KIT · CLAUDE CODE

§3 · MISSION TYPES

Three shapes a mission can take.

Pick the one that matches what's already true about your team and your timeline.

[01] MISSION2 weeks

SPRINT

One spec, one shipped slice. For teams who need a vertical AI feature live — not a deck.

e.g. RAG-backed search across your docs, deployed to your stack.

[02] MISSION4–8 weeks

BUILD

Full product, zero-to-launch. We embed a small crew, write the spec with your founders, and ship behind a domain.

e.g. full AI-native B2B SaaS, with auth, billing, monitoring, evals.

[03] MISSIONongoing

EMBED

A senior AI engineer on your team's payroll cadence, running on sunset. Monthly retainer.

e.g. you have a team, you need someone who's already shipped AI products to lead the AI surface.

Scope a mission

hello@bysunsetai.com

§4 · NEXT · 2026

Where sunset is going.

Today, sunset is the framework we use to ship missions. Soon, it's the OS we run our agency on — and that other agencies will too.

Agents that page you when a prod eval drops. VMs that spin up for every PR. CI/CD that knows the shape of your AI. Monitoring that calls your phone when latency tail breaks.

Right now we're building it for ourselves. We'll open it when it's real.

Join the alpha

Available soon — closed alpha.

ROADMAP · N3 LIVE
  • Spec-driven scaffoldshipped
  • Eight-decision provisionershipped
  • Mission VMs — ephemeral per PRalpha · 2026
  • Eval-aware CIalpha · 2026
  • Agent calls — voice updatesalpha · 2026
  • Monitoring with on-call routingalpha · 2026

REV. C · 2 SHIPPED / 4 ALPHA

§5 · EXAMINATION RECORD

Questions on file.

What buyers ask before they sign. Answered on the record.

[Q.01]

What is SunsetAI?

SunsetAI is an AI-native engineering agency: you bring a problem, we sign a spec, and we ship a production AI product in weeks. Every mission runs on sunset, the open-source framework we built and maintain. The crew previously shipped at Meta, AssemblyAI, CrowdStrike, and INRIA.

[Q.02]

How can you ship a production AI product in weeks?

Because the rails already exist. Every mission starts on sunset, where eight production decisions — cloud, database, auth, messaging, secrets, infrastructure-as-code, CI/CD, LLM routing — are pinned before any code is written. AI writes the how against a signed spec; humans control the what. A SPRINT ships in 2 weeks, a full BUILD in 4 to 8.

[Q.03]

What is the sunset framework?

sunset is an MIT-licensed, spec-driven framework for shipping production AI applications. It scaffolds a deployable system with Cloud Run, Postgres with migrations, JWT auth, Pub/Sub, Secret Manager, Terraform, GitHub Actions, and a multi-provider LLM router with pgvector retrieval. It is public: any engineer can audit what your product will run on before you hire us.

[Q.04]

How is a mission priced?

Per shape, not per hour. SPRINT (2 weeks) and BUILD (4–8 weeks) are fixed-scope: the price is written into the spec, and the spec is signed before code. EMBED is a monthly retainer for a senior AI engineer inside your team. You never receive an hourly invoice; you receive a signed spec with a number on it.

[Q.05]

Who owns the code and the IP?

You do. Everything built during your mission lands in your repository, your cloud account, your domain. The framework underneath is MIT-licensed, so there is no vendor lock-in: any engineering team can run, audit, and extend what we ship without us.

[Q.06]

How do you handle data privacy and security?

Your data stays in your infrastructure. Missions deploy into your own cloud account via Terraform, secrets live in your secret manager, and we never train models on your data. The crew includes engineers who shipped security products at CrowdStrike; GDPR-sensitive deployments are routine, not an exception.

[Q.07]

What if AI is the wrong tool for our problem?

Then we say so on the first call. Some problems need a database and a form, not a model — and a spec that can't define a measurable evaluation doesn't become a mission. We'd rather scope you a smaller, correct project than sell you an AI system you don't need.

[Q.08]

How does a mission start?

Email hello@bysunsetai.com. We do a 30-minute scoping call, then write a spec that fixes scope, price, and evaluation criteria. You sign it, the crew starts, and the first deploy typically lands within the first week.