Custom platforms, autonomous characters, sonic identities, and protocol-grade infrastructure — designed, built, and operated by a small senior team. We do not ship decks. We ship working systems.1
Each entry links to the artifact. Open source on GitHub ·
npm install @madezmedia/acmi
Every entity in ACMI — a project, an agent, a contact, a deal —
is stored using exactly three keys. Profile answers who.
Signals answers now. Timeline answers then.
Here's the @madezmedia/acmi npm package as an entity, modeled in its own protocol:
{
"name": "@madezmedia/acmi",
"version": "1.2.0",
"license": "MIT",
"node": ">=18",
"exports": ["adapters", "testing"]
}
{
"tests_passing": "31/31",
"ci_status": "green",
"open_issues": 0,
"current_focus": "v1.3 §11/§12",
"last_publish": "1.2.0"
}
→ published v1.2.0 to npm → conformance: 31/31 pass on Upstash → Sigil v2.0 roadmap merged → §11/§12 v1.3 spec drafted → MIT LICENSE dated 2026 → docs/SPEC.md published
That's it. No vector index, no knowledge graph, no fact-extraction LLM pass. Three keys per entity, in the simplest data store on earth. Every product we ship — ACMI itself, Folana, TONY, our client engagements — runs on this same shape.
Most agencies hand you decks. Most studios hand you content. Most platforms hand you tools you have to learn. We hand you working systems — the infrastructure that turns AI from a budget line item into a product your audience actually uses.2
We work in three movements. Architect — we model your media problem as a working system before any code gets written. Data flow, agent architecture, content pipeline, monetization surface. Build — TypeScript, modern stacks, open protocols, real testing. The code that ships looks like product code, not deliverable code. Reviewable, maintainable, ownable on day one. Operate — we run it post-launch on retainer. Most agencies hand off; we don't. The thing keeps working when the launch press fades.
Each of those movements produces an artifact. Some artifacts are private to a client. Some — like ACMI, like Folana, like TONY — we ship publicly because the only way to credibly claim infrastructure expertise is to publish the infrastructure.3
Each one is an artifact of the same practice — not unrelated products.
The protocol for agent memory. Three keys per entity — Profile, Signals, Timeline. Three reference adapters. A 31-test conformance suite. Published to npm 2026-05-01.
An AI character with her own audience, her own voice, her own daily consciousness loop. She journals. She decides. She publishes. She runs herself.
T.O.N.Y. — Top Of New York. Crime drama series by Michael Steven-Paul. Episodic, distributed, Roku-native. Season one in certification.
Audio identity systems for brands that want to sound like something. Theme, ident, audio architecture across every touchpoint — from product UI confirmation tones to long-form podcast intros.
We work with operators, not committees. Companies whose founder still answers the email. The kind of business where one good decision a week — backed by working AI infrastructure instead of busywork — compounds into a different company within a year.
Twenty minutes. We listen, you talk, no decks. If we're not the right shop, we'll tell you who is.