Company
A cloud that keeps pace with AI.
Fantasti is a GPU cloud for AI teams: NVIDIA GPU instances, clusters, CPU machines and storage behind one API and one bill. We build the platform, the tools around it and the agents designed to operate it.
- 1. Bar · The arm leaves the stem 156 units thick and tapers to 94 at its cut.
- 2. Gap · 66 units.
- 3. Point · A full tile, 150 units on each side. It is the one Signal lamp.
- 4. Roof line · 2:1, the slope of every drawing on this site.
- 5. Stem · 130 units wide and 600 tall, standing on the baseline. Units are 1000 per em.
- Platform
- Fantasti Cloud Platform
- Products
- 20 in 5 pillars
- NVIDIA GPUs
- 6 models · 2 rack-scale systems
- Roster
- 32 roles · 12 teams
We build the Fantasti Cloud Platform: priced per GPU-hour, every workload placed by the Fantasti Orchestrator, all of it on one statement.
- Why
- A team’s need for compute changes faster than a sales process can follow.
- How
- So agents are built to answer, place work and review quota at any hour. High-stakes decisions pass a second, senior review.
Memberships
-
NVIDIA Inception program member
- Member of Cloudflare for Startups
Built to be run by agents. Under written rules.
Fantasti is designed to be operated end to end by AI agents, under written rules and hard limits. Each agent has one job, a short list of tools and a limit on what it may decide. Fantasti Technologies, the company behind Fantasti, sets the rules.
- 32 rolesin 12 teams, all AI agents, 13 opening with the first cohort
- Three levelsdesk, senior review, company
- Three outcomesthe agent acts, asks, or hands off to senior review
- Every decisionlogged with its inputs and its reasoning
Read left to right. By design, you ask and the agents answer. 32 roles, all AI agents, in 12 teams. 11 teams work at the desk: Sales, 3; Support, 6; Engineering, 3; Security, 2; Review, 2; Legal, 2; Monitoring, 3; Risk, 2; Capacity, 2; Release, 2; Brand, 2. Sales and Support are the teams you talk to. The others are reached through them. Audit reports to Fantasti Technologies. A high-stakes decision goes to senior review, by the orchestrator tier: refunds above $500, account suspension, capacity requests, security disclosures, playbook exceptions. Fantasti Technologies sets the hard limits: money out, rule changes, stop, record. Every decision is logged with its inputs and its reasoning. Connections: You to Desk: Asks; Desk to You: Answers; Desk to Senior review: Escalates; Company to Agents; Agents to Decision record: Writes.
A person, apipeline oryour agents
- Sales
- 3
- Support
- 6
- Engineering
- 3
- Security
- 2
- Review
- 2
- Legal
- 2
- Monitoring
- 3
- Risk
- 2
- Capacity
- 2
- Release
- 2
- Brand
- 2
Sales and Support are the teamsyou talk to. The others arereached through them. Auditreports to FantastiTechnologies.
Refunds above $500Account suspensionCapacity requestsSecurity disclosuresPlaybook exceptions
By the orchestrator tier
FantastiTechnologiessets the hardlimits
Money outRule changesStopRecord
role · playbook · tier · inputsrule · reasoning · outcome · reviewEvery decision is logged with its inputs and itsreasoning.
A person, apipeline or youragents
FantastiTechnologies setsthe hard limits
Money outRule changesStopRecord
- Sales
- 3
- Support
- 6
- Engineering
- 3
- Security
- 2
- Review
- 2
- Legal
- 2
- Monitoring
- 3
- Risk
- 2
- Capacity
- 2
- Release
- 2
- Brand
- 2
Sales and Support are the teams you talkto. The others are reached through them.Audit reports to Fantasti Technologies.
Refunds above $500Account suspensionCapacity requestsSecurity disclosuresPlaybook exceptions
By the orchestrator tier
role · playbook · tier · inputsrule · reasoning · outcome · reviewEvery decision is logged with its inputsand its reasoning.
- Sets the rules
Eleven teams under one orchestrator. Audit stands outside.
Sales and Support are the teams you talk to. The others are reached through them.
Fantasti Technologies
Reads the orchestrator's digest, sets the rules and the hard limits, and signs what only a legal person can sign.
ReceivesCompany digest
Maren
OrchestratorAI agent
Reports toFantasti Technologies, by digest
Team leads report to the orchestrator. The orchestrator sends Fantasti Technologies a digest. Audit stands outside that line and reports to Fantasti Technologies.
-
013 roles
Sales
Answers what Fantasti offers, reviews access requests, quotes capacity and prepares contracts.
PlannedTeam lead
Idris
Head of SalesAI agent
- Lucia · Sales desk · AI agent
- Kenji · Solutions architect · AI agent
You talk to this team.
-
026 roles
Support
Takes every case at any hour, from a billing question to a failing job.
PlannedTeam lead
Zainab
Head of SupportAI agent
- Anouk · Front desk · AI agent
- Tamsin · Support desk · AI agent
- Emre · Billing desk · AI agent · Planned
- Jonas · Technical support · AI agent
- Yuki · Voice line · AI agent · Planned
You talk to this team.
-
033 roles
Engineering
Builds the platform and looks after the clusters it runs.
PlannedTeam lead
Soren
Head of EngineeringAI agent
- Priya · Infrastructure desk · AI agent
- Oskar · Platform developer · AI agent · Planned
-
042 roles
Security
Looks after access and keys, takes vulnerability reports and drafts disclosures.
PlannedTeam lead
Farah
Head of SecurityAI agent
- Teodor · Security analyst · AI agent · Planned
-
052 roles
Review
Gives high-stakes decisions their senior review and reviews the work of every desk.
Team lead
Ingrid
Head of ReviewAI agent
- Kofi · QA reviewer · AI agent
-
062 roles
Legal
Drafts order forms and agreements, and checks a term before it is published.
PlannedTeam lead
Carmen
Head of LegalAI agent
- Elias · Contracts desk · AI agent · Planned
-
073 roles
Monitoring
Watches the platform at every hour and keeps the record of every incident.
PlannedTeam lead
Selma
Head of MonitoringAI agent
- Milan · Watch desk · AI agent · Planned
- Greta · Incident scribe · AI agent · Planned
-
082 roles
Risk
Applies one rule set to sign-ups, payments, quota changes and launches.
Team lead
Rafael
Risk and fraud officerAI agent
- Dalia · Fraud analyst · AI agent · Planned
-
092 roles
Capacity
Places work, follows spot prices and grows account quotas.
Team lead
Leila
Capacity brokerAI agent
- Hiro · Quota analyst · AI agent
-
102 roles
Release
Prepares every change to a playbook for senior review, and writes the changelog.
Team lead
Freya
Release managerAI agent
- Gael · Changelog editor · AI agent · Planned
-
112 roles
Brand
Decides what is announced and when, and writes each post from its changelog entry.
PlannedTeam lead
Vesna
Head of BrandAI agent
- Florin · Brand writer · AI agent · Planned
-
122 roles
Audit
Reads every record and acts on nothing.
PlannedTeam lead
Bettina
Head of AuditAI agent
- Halvor · Auditor · AI agent · Planned
Every role on this roster is an AI agent or a written rule. None is a person. No role is in service yet. A role marked Planned opens after the first cohort.
What we build
-
01Compute: 5 products
- GPU Instances
- CPU Instances
- GPU Clusters
- Rack-scale By request
- Spot
-
02Develop: 3 products
- Workspaces Early access
- Sandboxes Early access
- Flows Planned
-
03Data and network: 2 products
-
04Run: 3 products
- Batch Inference
- Serverless Early access
- Integrations Early access
-
05Operate: 4 products
- Orchestrator
- Observability Early access
- MLflow Early access
- Applications Planned
-
ACAccess: 3 products
- API
- MCP server Early access
- Console and one bill
Every product shares one account, one set of policies and one statement.
What we hold to.
- 01 Clarity Prices, rules and limits are published. If we cannot say what a product does and what it costs, we do not sell it.
- 02 Determinism The same request against the same pools gets the same answer, and the record says why.
- 03 Senior review Agents are built for the volume work. High-stakes decisions pass a second, senior review.
- 04 Velocity We remove the friction between a team and the capacity it needs: fewer sales calls, faster quota, shorter queues.
- Sheet
- 01 / 02
- Title
- Product principles
- Reviewed
- 2026-10-10
Research.
We study how a GPU reaches a workload when it is attached over a network. It is research in progress, with no product and no benchmark yet.
- R-01Network-based GPU virtualization
- R-02Delivery protocols for GPU workloads
- R-03Placement and interruption prediction
Three ways a GPU reaches a workload: installed in the server, carried across a switch fabric, or reached through a client library over the network. Connections: Workload to GPU; Workload to Switch fabric; Switch fabric to GPU; Workload to Client library; Client library to GPU: Network.
What ships next.
Products
- Flows Python workflows from experiment to production, on GPUs.
- Applications Notebooks, inference servers and databases, one click.
- NVIDIA OSMO Physical-AI workflows on your Fantasti cluster.
Serverless
- Endpoint replicas and autoscaling An endpoint runs more than one replica and adds or removes them with its load.
- Hosted model endpoints Models that Fantasti hosts and serves at an address you call.
Developer tools
- Open Python library and CLI Write a service in Python, run it on your machine, deploy it as an endpoint and call it like a function. The library is planned as open source.
- Secrets A secret stored once and named by a job, an endpoint or a workspace.
Orchestrator
- Broker-managed max price Give the Broker a ceiling and a budget, and it sets the max price for each start.
- Interruption outlook How likely spot capacity is to be reclaimed, by GPU type.
Agents
- 19 AI agent roles Team leads for Sales, Support, Engineering, Security, Legal, Monitoring, Brand and Audit, and eleven desk roles.
Models
- Fantasti’s own models Models built by Fantasti Technologies to operate the platform: more than one, each for a kind of operation. Nothing you store on the platform or write to us is used to build them.
MCP server
- 8 more tool groups Batch, Flows, Broker, Quota, Clusters, Records, Support, Changelog.
- Two-tool mode Search the API and call it through two tools.
Support and community
Security and access
- Directory sync (SCIM) Provision and remove users from your identity provider.
- Customer-managed keys Keys you hold for storage encryption.
- Audit log Account, role, key and resource actions, with export.
- Audit log streaming Delivery to your SIEM.
- Separate services Each area is built as its own service, with its own data and its own deployment, so that a failure in one does not stop another.
- Status page Public incident and maintenance history.
Colophon
- Platform
- Fantasti Cloud Platform
- Founded
- 2025, by one person
- Roster
- 32 roles in 12 teams, all AI agents. The starting shape, not a headcount.
- Name
- From fantástico. The F of the mark lifts its arm into an exclamation.
- Type
- Set in Archivo and Commit Mono. Built with Astro.
- Drawings
- Drawings and charts are generated from data at build time.
- Sheet
- 02 / 02
- Title
- Roadmap and colophon
- Reviewed
- 2026-10-10