The runtime layer

The Infrastructure
Behind Enterprise AI.

Carrier OS isn’t another AI application — it’s the runtime layer enterprises run their AI on. It holds a persistent, governed Truth State, so models stop losing context and every execution stays verified and auditable.

Read the thesis
01/ The Problem

AI rebuilds understanding every prompt.

Model access alone doesn't make AI reliable. Understanding is thrown away and reconstructed on every request — Carrier maintains it.

session · statelessTurn 42
usrShip to Berlin — net-30, invoiced in EUR.dropped
· · ·41 turns later
usrWhich currency did we agree on?
aiWhich currency would you like to use?
Context discarded — rebuilt from the prompt
session · governed truth stateTurn 42
usrShip to Berlin — net-30, invoiced in EUR.held
· · ·41 turns later
usrWhich currency did we agree on?
aiEUR — set at turn 1, held in the Truth State.
State held — verified against turn 1
What holds it together Truth State Compiler Persistent runtime Execution verified Governed by design Long-context stable
02/ The Runtime

Evidence in. Truth out.

Company knowledge becomes reusable operating state — compiled once, executed reliably, and evolved under governance.

The Carrier runtime loop

01

Evidence

Enterprise sources gathered

02

Structured State

Governed, versioned truth

03

Runtime Package

Execution-ready, sealed

04

Agent Execution

Runs from approved truth

05

Traceable Output

Every run leaves a receipt

06

Patch Evolution

Approved changes compound

Every run leaves state the next run can build on.

03/ The Proof

The runtime, measured.

Not projections — internal A/B benchmarks of the Carrier runtime against a stateless baseline.

3–4×

Higher information density

vs baseline

More useful knowledge stays active in the model's working state, so reasoning stays focused as information sets grow.

Lower knowledge growth

50–70%

Lower knowledge growth

vs baseline

Information stays organized and reusable instead of being repeatedly reconstructed — less context bloat, same continuity.

Faster response time

83%

Faster response time

vs baseline

Less compute spent recovering context means more compute for useful inference, across extended multi-turn sessions.

Stable extended inference

500K+

Stable extended inference

vs baseline

Coherent execution across 500K+ token workloads and continuous reasoning sessions approaching 28 minutes.

A/B tested across extended multi-turn sessions (~240 prompts) and 20MB+ information sets.

04/ Business Model

One runtime, many revenue lines.

The Website Builder is the wedge; the runtime underneath monetizes across every surface it powers.

Website Builder SaaS

Website Builder SaaS

Subscription access to the first Carrier product surface — the visible proof of the runtime.

Runtime API

Runtime API

Pay-per-execution access to the runtime for developers and internal builders.

Enterprise Licensing

Enterprise Licensing

Self-hosted, governed deployment for large organizations with their own data.

Marketplace

Marketplace

A take rate on Truth State assets, agents and connectors created, shared and sold.

Professional Services

Professional Services

Implementation, integration and enterprise onboarding for high-value accounts.

05/ Growth Flywheel

Compounding value, network effect by design.

Every layer feeds the next — more usage makes the runtime more valuable.

01 Developers

Build on Carrier OS

02 Applications

More powerful AI apps

03 Organizations

Adopt across teams

04 Executions

More runs, more context

05 More builders

Better tools attract devs

06/ Roadmap

From wedge to platform.

A staged path — the chat platform and organization workspaces are live today. Next comes the state layer, then the agent runtime on top of it.

Phase 1

Multi-Model Chat Platform

Live: streaming multi-model chat, document grounding, tool execution, and metered per-request billing.

Phase 2

Organization Workspaces

Live: organization accounts, role-based access, seat billing, member invitations, and separate personal and team workspaces.

Phase 3

State Layer Foundation

Store structured project, brand, website, page and asset state, with history and missing-information prompts.

Phase 4

Website Builder

MVP in private beta: project creation, style intake, page planning, templates, and generated React output. Hardening toward general release.

Phase 5

Agent & Runtime Layer

Task routing, Website/Content/Design agents, execution receipts, patch suggestions, and approval flow.

Phase 6

Platform Expansion

Document Generator, Marketing Studio, developer exports, API foundations, and early pilots.

Phase 7

Enterprise Hardening

Audit logs, compliance-ready logging, SSO, and deeper governance controls on top of the workspaces already shipped.

07/ Where It Applies

Built for real company work.

The same governed runtime is built for the highest-value, longest-horizon enterprise workloads.

01

Enterprise AI Assistants

Company-wide copilots grounded in real state.

02

Multi-Agent Orchestration

Coordinated agents sharing one truth.

03

Software Engineering

Long-running codebase reasoning.

04

Legal Analysis

Contract and clause review at depth.

05

Financial Research

Multi-document analysis that holds context.

06

Scientific Computing

Extended reasoning over large datasets.

07

Healthcare Workflows

Governed, auditable clinical support.

08

Long-Running Autonomous Systems

Stable execution across long horizons.

08/ Investment Thesis

Why we believe this is inevitable.

The runtime layer is the next foundational category in enterprise software.

01

Why now

AI shouldn't rebuild understanding every prompt. As it moves into mission-critical work, enterprises need reliability, governance and long-horizon memory — now.

02

Why Carrier

Foundation models reason; Carrier manages the operating environment. The long-term asset is the Truth State, not the prompt.

03

Why inevitable

Every serious AI application will need a runtime layer above the model. Carrier becomes the standard place company truth lives and executes.

09/ FAQ

Questions investors ask.

The answers behind the thesis.

RAG retrieves relevant information for a single prompt. Carrier structures, versions, governs, and reuses company knowledge as persistent operating state across many tasks.

Frameworks help agents run. Carrier gives agents shared state, structured context, approval flows, and traceable execution — so agents don't each invent their own version of the company.

Websites are visual, easy to demo, and depend on many forms of company truth — brand, audience, product, messaging, design, proof, and conversion goals.

Yes. Foundation models are replaceable reasoning engines. Carrier's value is the runtime layer: state, context preparation, agent coordination, execution tracking, and governance.

The structured state layer and runtime architecture. As users build more state inside Carrier, the system becomes more useful, reusable, and harder to replace.