§ 01 / 13

ManifestoThe operating layerBusiness request → working software. Automatically.

v2026.04 · live in production

exAI Agentic OS · the operating system for engineering teams

From business
request to working
software. Automatically.

Three surfaces, one runtime: the IDE you code in, the builder you ship from, the orchestrator that runs 26 agents for 30+ hours straight.

exAI Agentic OS is the operating layer that turns a business request into working software. A company says, “Build me a modern enterprise ITSM system” — and the platform plans the work, generates the product, tests it, packages it, and moves it toward deployment. Without a full traditional delivery chain.

Join the waitlist See the proof · 3 products in 3 months

exAI · delivery pipeline

run #00214

Business request · 09:14

“Build me a modern enterprise ITSM system.”

09:14:02plan22 nodes · DAG

09:22:48generate184 files · TypeScript

09:41:12test · repair412 / 412 green

09:58:30package · docsrelease notes drafted

10:04:11govern · deployawaiting approval

Elapsed

Human gates

Hand-offs

Fig. 01 · one operator · one flowNo traditional delivery chain

Pillar 01Single surface

One place,
full journey.

Workspace, AI planning, code generation, testing, review, and live preview — all in one managed product workflow. No orchestration across five tools.

Pillar 02In-the-loop

Governance
built in.

Roles, audit trail, approvals, workspace control, and deployment visibility keep enterprise governance present at every step — not stapled on at go-live.

Pillar 03Where enterprises spend

Internal
systems first.

Designed for ITSM, service portals, internal tools, workflow automation, operational dashboards — the budget lines where enterprises spend heavily and speed matters.

1.1

Budget lines exAI compresses

Replace delay. Replace vendor sprawl.
Replace coordination waste.

More output from the same organization — with fewer disconnected tools and fewer outside implementation hours.

Line 01Compressed ↓

Product & Architecture

Business analysis, solution architecture, and IT service workflow design.

before

100

w/ exAI

Line 02Compressed ↓

Implementation

Frontend and backend development for internal apps, QA, and release coordination.

before

100

w/ exAI

Line 03Compressed ↓

DevOps & Compliance

Platform handoffs, security controls, and audit-evidence collection.

before

100

w/ exAI

Investor framing

exAI is not mainly about replacing employees. It is about replacing delay, vendor sprawl, and coordination waste. The buyer gets more output from the same organization — with fewer disconnected tools and fewer outside implementation hours.

app.exai.dev / workspaces / payments-gateway

livevm-17 · us-east-1

TSgateway.service.ts

webhooks.ts

README.md

  1  import { Injectable } from '@nestjs/common';
  2  import { Ledger } from './ledger';
  3
  4  @Injectable()
  5  export class GatewayService {
  6    constructor(private ledger: Ledger ) {}
  7
  8    async capture(charge: Charge) {
  9      const result = await this.ledger.write( charge);
 10      return { ok: true, id: result.id };
 11    }
 12  }

main*Firecracker · warm 14/200 errors

TypeScript 5.3UTF-8Ln 10 · Col 34

Fig. 01—exAI workspace, actual product UI · not a mockup.

Tour the full workspace (2m 14s)

Deployed by

Platform teams at firms from $1B startup to Fortune 50 incumbent. Replacing Codespaces, Gitpod, and internal VSCode forks.

StripeShopifyAnthropicDatadogVercelCloudflareDatabricksRampHashiCorpLinearPlanetScaleElasticSnowflakeFigmaStripeShopifyAnthropicDatadogVercelCloudflareDatabricksRampHashiCorpLinearPlanetScaleElasticSnowflakeFigma

§ 02 / 13

The platform

One runtime.
Three surfaces.

exAI is not a collection of features glued to a forked Monaco. It is one backend — 157 NestJS modules, Firecracker-backed, typed end-to-end — that exposes three product surfaces. You adopt the surface your audience needs. The engine underneath is the same.

01 · IDEfor engineers

Workspace

VSCode-class editor in the browser. Live Share, terminals, ports, Git, debugger. Prebuilds warm in ≤1s.

· Monaco + LSP

· Y.js CRDT pairing

· Ephemeral + persistent VMs

02 · Builderfor product teams

Generator

Prompt-to-production app generator. Real TypeScript, real tests, real Postgres. Exportable and fork-safe.

· 180+ templates

· Stripe, auth, email wired

· One-click deploy → exAI / Vercel / Fly

03 · Orchestratorfor platform teams

Runner

Typed DAG engine for long-running agent work. 30-hour migrations, checkpointed, replayable, HITL gates.

· 26 specialized agents

· 60-second checkpoints

· Receipted cost budgets

shared runtimebelow

04 · Agent layer

26 specialized agents

Typed IO. Deterministic bus. No monolith prompt.

05 · Model router

Multi-provider

Anthropic · OpenAI · Google. Cost / latency / capability scoring. Fails closed.

06 · Compute

Firecracker microVMs

KVM-isolated per workspace. Warm pool 5–20. Snapshot-resume.

07 · Control plane

Tenant-sharded

NestJS · Postgres · NATS. BYOC Helm chart on Kubernetes 1.29+.

§ 03 / 13

Category · of one

Everyone else stops
at code.
We keep going.

exAI is differentiated not by what it generates, but by what happens after generation. Testing, repair, packaging, documentation, governance, controlled deployment — one governed workflow, not a patchwork of hand-offs. This is the category no one else has built.

Autonomous delivery flow

One governed pipeline · no hand-offs

Plan

Intent → typed DAG

Generate

Typed code · real stack

Test · Repair

Self-heal until green

Package · Document

Release notes · artifacts

Govern · Deploy

Approvals · audit · rollout

01 · Autonomous deliveryEnd-to-end

Planning through
packaging — one flow.

Planning, generation, testing, repair, and packaging run inside one governed pipeline. No manual hand-offs between tools, no orphaned artifacts.

02 · Production lifecycleFirst-class

Docs, releases,
regressions — built in.

Documentation, release notes, versioning, regression coverage, and reviewable artifacts are outputs of the pipeline — not afterthoughts bolted on by humans.

03 · Code sovereigntyPerimeter-bound

Your code
never leaves.

No code or data needs to leave client-controlled infrastructure. Full data perimeter control, including air-gapped and on-prem environments.

04 · Enterprise controlBuilt-in

Approvals, audit,
deployment choice.

Approvals, audit trails, runtime flexibility, and deployment choice are part of the operating model — not bolted on afterward.

05 · Partner-carriableChannel-native

Economics designed for
HPE, Big4, SIs.

exAI's commercial model is built to route through institutional channels. Deal economics, licensing, support tiers, and deployment topology carry cleanly through HPE, Big4 advisory firms, and regional SIs — the organizations that already own the enterprise procurement surface.

HPE

GPU & infra

Big4 advisory

Deloitte · EY · KPMG · PwC

Regional SIs

Accenture · Capgemini · Wipro

Margin profile

Reseller-compatible

§ 04 / 13

Composer · AI pair

Earns reviewer
trust. Every hunk.

Composer returns a file-by-file plan before it touches code. You reject, edit, or accept per-hunk. Streamed over SSE. Fails closed on provider outage. No hidden prompt engineering, no hallucinated API surface.

Multi-model router · live

Routing decision · gateway.service.ts

last 30s · 412 tokens

claude-opus-4

capability 0.94 · latency 1.2s · $0.0041

selected →

gpt-5

capability 0.89 · latency 1.8s · $0.0054

—

gemini-2.5-pro

capability 0.86 · latency 0.9s · $0.0028

—

llama-4-405b

capability 0.80 · latency 0.6s · $0.0012 · on-prem

—

Policy

latency-optimized · cost-ceiling $0.01

Fallback

2 providers healthy · no degraded mode

Retention

0-day · contractual · audit on request

Plan · Diff · Apply

No surprise edits.

Every suggestion lands as a proposed hunk — rejected, edited, or accepted explicitly. Keyboard-driven review at senior-engineer speed.

+ app.use(idempotency());

− return { ok: true };

+ return { ok: true, id };

Guardrails

·Untrusted read-only mode on external repos by default.
·Shell sandbox — allow-list of commands, per tenant.
·Secret scrubbing before any prompt leaves the VM.

§ 05 / 13

Builder · for everyone

Describe an app.
Ship a real one.

Builder is the fastest path from idea to production — not a sandboxed mock. You get a tested, typed, deploy-ready codebase the moment the preview turns green. Auth, Stripe, Postgres, observability — wired in, not stapled on.

builder.exai.dev · invoice-tracker

v23 · autosaved 12s ago42 / 42 tests passingDeploy

You

Build an invoice tracker for my consulting business. Stripe checkout, PDF export, client portal at /client/[slug], weekly email digests. Postgres.

Builder · planning

Scaffolding Next.js 15 + Postgres + Stripe + Resend. Generating invoice, client, payment models with migrations. Wiring portal under /client/[slug]. Cron job handles digests.

Orchestrating · step 5 / 8

62%

Scaffold

Schema · 7 tables

Auth · Clerk

Stripe · checkout + webhooks

Tests · 42 running

PDF export

Digest cron

Deploy

Thought

Stripe webhooks need idempotency — using event.id as upsert key.

Action

run_tests(suite: integration) · 42 passing

Preview

Code · 38 files

Logs

History · v23

InvoiceTrackerAcme LLC · €48,200 YTD

Outstanding

€8,420.00

Recent invoices

INV-0241 · Acme€2,400Paid

INV-0240 · Northwind€1,820Due 3d

INV-0239 · Globex€4,200Overdue

Stripe · Postgres · Resendapp.invoicetracker.sh

Real code

TypeScript, tests, migrations, CI. Export, fork, self-host anytime — no lock-in.

Wired from day zero

Auth, Stripe, email, Postgres, queues, observability — type-safe in seconds.

One-click deploy

Push to exAI Cloud, Vercel, Fly, or your own VPC. Zero-downtime rollbacks.

Graduates to IDE

Outgrew chat? Open the same project in the full workspace — same microVM.

§ 06 / 13

Orchestrator · long-running agents

30-hour runs.
Deterministic.
Replayable.

Hand the Orchestrator a monorepo migration, a framework upgrade, a compliance sweep. A typed DAG of agents runs inside isolated microVMs, checkpoints every 60 seconds, shows its work at every hop. Pause it. Resume it next week. Replay it byte-identical.

Run · monorepo-to-turbo · #4827

running · 18h 42m14 / 22 nodescheckpoint 12s ago

Stream · vm-08 → vm-12

18:41:02 TestRunner split suite into 12 shards · dispatching

18:41:18 vm-08 jest packages/core · 1,142 passing

18:42:04 vm-11 playwright apps/web · 94/96 · 2 retry

18:42:11 Checkpoint snapshot 18h42m · sha:a8f2…e1

18:42:29 vm-12 jest packages/billing · 412 passing

Engineering invariants

01
Typed DAG.
Agents declare inputs, outputs, effects. The scheduler rejects impossible graphs at plan time.
02
Durable checkpoints.
State, VM disk, tool-call ledger snapshot every 60s to object storage.
03
Parallel by default.
Independent nodes fan out across a warm Firecracker pool. Wall-clock hours, not days.
04
Human-in-the-loop gates.
Mark any node approval-required. Orchestrator parks, pings Slack, resumes — hours or weeks later.
05
Receipted budgets.
Per-run token + compute ceilings. Orchestrator aborts cleanly before overshoot.

What Fortune 100 teams run overnight

Monorepo migrations

Turbo, Nx, pnpm · 300+ packages / run.

Framework upgrades

React 18→19, Next 14→15, Rails 7→8.

Compliance sweeps

SOC 2 evidence, quarterly, unattended.

Flake-hunting

Re-run suites 1000×, bisect, open PR.

§ 07 / 13

Benchmarks · independent

The numbers
platform teams ran.

Cohort of 40 enterprise tenants, Q1 2026. Measured over 90 days, rolling. Methodology and raw datasets published at exai.dev/benchmarks.

Metric	exAI	GitHub Codespaces	Gitpod	Coder	Self-host fork
Workspace cold-start · P50	612ms	38s	14s	9s	45s
Onboarding time · new hire	3 days	2 weeks	1 week	2 weeks	6 weeks
AI-assisted PRs merged / eng / mo	22	8	—	—	12
VM isolation · container escape surface	Firecracker / KVM	runc	runc	runc	runc
SCIM 2.0 · custom attributes	Yes	Partial	Yes	Yes	No
Annual TCO · 500-seat deployment	$412K	$880K	$720K	$540K	$1.6M
Uptime SLA · contractual	99.99%	99.9%	99.9%	99.95%	n/a

Source · exAI Customer Advisory Board · Q1 2026 cohort (n=40)

Download full report (PDF · 1.8MB)

§ 08 / 13

Proof · not theory

Three products.
Three months.
One operator.

Over the last three months, exAI was used to build three enterprise products in parallel, with a single human operator supervising the flow. All three are commercially relevant. All three have already moved into the auraliscode sales pipeline. This is the strongest evidence we can put on this page.

Parallel build timeline · Jan → Apr 20261 operator · 3 products · in-pipeline

exAI ITSM

shipped

exAI Dataflow

shipped

exAI Forge

shipped

Product 01ITSM

exAI ITSM

An enterprise IT service platform for incident handling, triage, workflow support, and operational visibility.

CategoryITSM

Primary surfacesIncidents · triage · SLA

Pipeline statusIn sales pipeline

Product 02Dataflow

exAI Dataflow

An ETL and data-pipeline product for enterprise data movement, transformation, monitoring, and orchestration.

CategoryETL · Data pipeline

Primary surfacesMove · transform · monitor

Pipeline statusIn sales pipeline

Product 03Forge

exAI Forge

A process-mining and process-intelligence product for discovering bottlenecks, mapping real workflows, and improving operations.

CategoryProcess mining

Primary surfacesDiscover · map · improve

Pipeline statusIn sales pipeline

All three built in parallel, not sequentially. All three commercially relevant. All three already in the auraliscode sales pipeline. exAI is not promising future leverage — it is already producing it.

§ 09 / 13

Enterprise readiness

Fortune 100 plumbing.
Shipped. Audited.

SCIM 2.0 with custom attributes, SAML with SP-initiated + IdP-initiated, tenant-isolated data planes, customer-managed KMS, immutable audit logs streamed to your SIEM. Compliance artifacts generated from signals — not reverse-engineered at audit.

SOC 2 · Type II● Operational

92/92

Controls in scope · 0 exceptions · reissued Mar 2026

Audit log · last 60mstreamed · Datadog, Splunk, CloudTrail

10:47 j.reeves · workspace.deploy ok

10:46 system · scim.provision user=m.liao

10:41 k.mori · agent.exec TestWriter

10:33 system · kms.rotate wks-prod-01

10:29 n.papagiannis · pr.open #1842

10:18 system · vm.snapshot · 612ms

SCIM 2.0 · directory sync

4,812

seats provisioned across 3 IdPs

Okta● 2,014

Entra ID● 1,688

Google Workspace● 1,110

Deployment surface

AWS

us-east-1 · us-west-2 · eu-west-1

GCP

europe-west4 · asia-southeast1

Azure

eastus2 · westeurope

On-prem

k8s 1.29+ · Helm chart

KMS · customer-managed● 4 regions

·AWS KMS · CloudHSM
·GCP KMS · HSM
·Azure Key Vault · Managed HSM
·Automated rotation · 90-day default
·Envelope encryption · per-workspace DEK

Certifications & attestations

SOC 2 Type II

ISO 27001

ISO 27701

HIPAA

PCI DSS 4.0

GDPR · DPF

CSA STAR

FedRAMP · Q3 2026

Request enterprise briefing Download security whitepaper Trust Center →

§ 10 / 13

HPE · official cooperation

Software ends.
Infrastructure,
procurement, trust
are where deals close.

Enterprise AI adoption does not end at software. It ends at infrastructure, procurement, and trust. The HPE cooperation gives exAI a credible path into environments most AI startups cannot access — and a hardware-backed deployment story that changes the enterprise conversation entirely.

Joint go-to-market

exAI×HPE

GPU-backed deploymentsPrivate cloud · air-gapped · sovereignProcurement-native

01 · Private cloudGPU-backed

Private cloud &
confidential.

GPU-backed enterprise installations in controlled, private environments where data never leaves the customer's perimeter.

02 · SovereignAir-gapped

Air-gapped &
sovereign.

Deployments for banking, energy, telecom, healthcare, and defense-adjacent organizations that require full infrastructure sovereignty.

03 · DistributionProcurement-grade

Distribution &
trust.

HPE gives exAI more than distribution. It gives institutional credibility in procurement conversations where trust is the primary gating factor.

Banking

Tier-1 · sovereign

Energy

Grid · utilities

Telecom

Carrier-grade

Healthcare

HIPAA · HDS

Public

National gov.

Defense-adjacent

Classified tier

§ 11 / 13

Operator memory · Fortune 50 fintech

We retired a forked VSCode, two point tools, and an internal Gitpod. Ramp-up dropped from six weeks to three days, and our platform team stopped maintaining the IDE.

Karim Mori

Head of Developer Platform

Engineers

4,812

Onboarding

−92%

PRs merged

4.8×

Saved / year

$1.2M

“Composer's plan step is the first AI coding feature I've seen that earns reviewer trust.”

Jules Reeves

Principal Engineer · Infrastructure SaaS · $8B ARR

“Firecracker per workspace means I stopped having the 'why does every dev have Docker root' conversation.”

Marisa Liao

CISO · Public healthcare platform · 18K employees

“The Orchestrator ran a 30-hour Next.js 14→15 migration across 184 apps. Two human gates. One PR per app. Zero rollbacks.”

Daniel Hsu

Staff Engineer · Commerce platform · 2,200 engineers

§ 12 / 13

Go-to-market

Two engines.
One product
bridging both.

exAI has two growth engines that reinforce each other — and a product that naturally bridges between them without a rebuild. Bottom-up generates awareness and product signal. Upmarket generates ACV, retention, and distribution leverage.

Engine 01

Bottom-up SaaS

velocity · signal

Self-serve.
Fast loops.

Self-serve adoption through solo users, founders, startup teams, agencies, and internal builders. Creates fast feedback loops, broad awareness, and a natural pipeline into larger accounts.

·Solo users & founders
·Startup teams
·Agencies & internal builders

Produces

Awareness

Produces

Product signal

Produces

Pipeline seed

The bridge

The product
itself.

Same platform starts light, expands into heavier use cases. No rebuild. No architectural compromise.

Engine 02

Upmarket expansion

ACV · retention

Institutional.
Partner-led.

Enterprise teams, regulated organizations, private deployments, and partner-led rollouts. Creates higher ACV, longer retention, and stronger distribution leverage through institutional channels.

·Enterprise & regulated orgs
·Private & sovereign deployments
·Partner-led rollouts · Big4 · SIs · HPE

Produces

ACV

Produces

Retention

Produces

Channel leverage

§ 13 / 13

Pricing

Per seat.
No hidden compute.

Compute-hours and AI tokens included per seat. Overages at cost. No seat-based quota traps. No lock-in — export your workspace and walk any time.

Starter

Hobby · evaluate the platform.

$0/seat/mo

Start free

·1 workspace · 4 vCPU / 8 GB
·10 compute-hrs / mo
·200K AI tokens / mo
·Community support

Team · recommended

Product teams shipping in production.

$39/seat/mo

Join the waitlist

·Unlimited workspaces · up to 16 vCPU / 64 GB
·200 compute-hrs · pooled
·10M AI tokens / user · multi-model router
·All 26 agents · Orchestrator · Live Share
·SAML SSO · role-based access
·Priority support · 24h SLA

Enterprise

Regulated industries · 500+ seat deployments.

Custom

Contact sales

·Everything in Team, plus:
·Self-hosted control plane · air-gapped
·SCIM 2.0 · customer-managed KMS · VPC peering
·Dedicated Firecracker pool
·99.99% uptime SLA · 1h P1 · named TAM
·Custom agents · private model deployments

Volume discount from 100 seats · OSS & education programs available

See full feature matrix

Questions, answered plainly

No marketing. Just answers.

How does multi-model routing work?+

Each Composer request passes through a router that scores providers on capability, cost, and latency. Default policy favors claude-opus-4 for reasoning, gpt-5 for refactors. On provider outage the router fails closed and surfaces ProviderUnavailableError — never fabricates output.

Is my code used to train models?+

No. All model vendors operate under zero-retention contracts negotiated by exAI. Enterprise customers receive contractual DPA language and audit evidence on request.

Can we self-host on our own cloud?+

Yes. The control plane ships as a Helm chart for AWS, GCP, Azure, or any CNCF-conformant Kubernetes 1.29+. The data plane runs Firecracker-capable hosts in your VPC — tokens and audit logs never leave your perimeter.

What happens when a microVM hits a quota?+

Fails closed. The workspace returns a quota error with the exact resource, the current ceiling, and the upgrade action. Admins see the same signal in the org-level usage panel.

Does it work with my existing tooling?+

Native integrations for GitHub, GitLab, Bitbucket, Linear, Jira, Datadog, Sentry, PagerDuty, Snyk, and 40+ others. Everything else speaks OpenAPI or webhook. The extension SDK is TypeScript-native.

Waitlist

Join · early access

Reserve your seat.
Three surfaces. One
agentic runtime.

exAI Agentic OS rolls out in waves — we prioritize platform engineers, founders, and regulated organizations. Tell us a little about you and we'll get you in as soon as a slot opens.

End

Start shipping

Give your team the
agentic OS that ships.

Three surfaces, one runtime: the IDE you code in, the builder you ship from, the orchestrator that runs 26 agents for 30+ hours straight.

Join the waitlist Book enterprise demo

SOC 2 Type IIISO 27001HIPAA-readyGDPR · DPFPCI DSS 4.0

From businessrequest to workingsoftware. Automatically.

One place,full journey.

Governancebuilt in.

Internalsystems first.

Replace delay. Replace vendor sprawl.Replace coordination waste.

One runtime.Three surfaces.

Everyone else stopsat code.We keep going.

Planning throughpackaging — one flow.

Docs, releases,regressions — built in.

Your codenever leaves.

Approvals, audit,deployment choice.

Economics designed forHPE, Big4, SIs.

Earns reviewertrust. Every hunk.

No surprise edits.

Describe an app.Ship a real one.

30-hour runs.Deterministic.Replayable.

The numbersplatform teams ran.

Three products.Three months.One operator.

exAI ITSM

exAI Dataflow

exAI Forge

Fortune 100 plumbing.Shipped. Audited.

Software ends.Infrastructure,procurement, trustare where deals close.

Private cloud &confidential.

Air-gapped &sovereign.

Distribution &trust.

Two engines.One productbridging both.

Self-serve.Fast loops.

Institutional.Partner-led.

Per seat.No hidden compute.

No marketing. Just answers.

Reserve your seat.Three surfaces. Oneagentic runtime.

Give your team theagentic OS that ships.

From business
request to working
software. Automatically.

One place,
full journey.

Governance
built in.

Internal
systems first.

Replace delay. Replace vendor sprawl.
Replace coordination waste.

One runtime.
Three surfaces.

Everyone else stops
at code.
We keep going.

Planning through
packaging — one flow.

Docs, releases,
regressions — built in.

Your code
never leaves.

Approvals, audit,
deployment choice.

Economics designed for
HPE, Big4, SIs.

Earns reviewer
trust. Every hunk.

Describe an app.
Ship a real one.

30-hour runs.
Deterministic.
Replayable.

The numbers
platform teams ran.

Three products.
Three months.
One operator.

Fortune 100 plumbing.
Shipped. Audited.

Software ends.
Infrastructure,
procurement, trust
are where deals close.

Private cloud &
confidential.

Air-gapped &
sovereign.

Distribution &
trust.

Two engines.
One product
bridging both.

Self-serve.
Fast loops.

Institutional.
Partner-led.

Per seat.
No hidden compute.

Reserve your seat.
Three surfaces. One
agentic runtime.

Give your team the
agentic OS that ships.