Cost to discover
Few shots on goal. Prototypes are expensive, so the decision to kill one is made politically, months after the data allowed it.
NewPrototype Machines — 1,000 shots on goal a year
The backend backbone for live game studios
Vlitz gives you production-grade backend systems out of the box — sessions, economy, inventory, progression, store, LiveOps — and supervised agents that build, deploy, and operate them beside your team.
Studios lose most of their money before the game is worth building
Every title rebuilds the same plumbing before a line of unique gameplay is written. And upstream of that, the money goes into games that should never have been built at all. Three costs dominate.
Time to launch
Months go into accounts, economy, inventory, progression and store tooling — none of it the thing that makes your game different.
Cost to operate
After launch, a standing team is required just to keep the game healthy: flags, tests, content pushes, incidents, deployments.
Vlitz attacks all three. The funnel widens discovery, prefab modules collapse build time, and supervised agents collapse the cost of operating what survives.
Prototype Machines
A thousand shots on goal, and a gate on every one
The portfolio playbook that built the hypercasual giants, rebuilt for midcore depth. Most ideas die at the first stage, on a full market test. That is the design, not the failure.
1,000Prototypes
- Gate
- Creative test, then a first market test
30Candidates
- Gate
- D0 playtime and sessions
6MVPs
- Gate
- D1 retention on a real cohort
3Soft launches
- Gate
- D3 and D7, with ROAS beside them
1Global
- Gate
- A validated payback curve
Gates are cumulative. Every gate re-checks all prior ones, so a game that regresses is caught rather than carried. Cost is measured, not guessed: Acquisition attributes each install to a channel, campaign and creative, then derives CPI against potential CPI — the most a game can afford to pay at its target payback. The gap between them is the honest scale-or-stop signal.
1
The machine recommends, humans decide
No agent promotes, demotes, or kills a game, releases money, or publishes to a store. Every such act is human, and carries a recorded rationale.
2
The threshold precedes the data
Gates are configured before the cohort exists. A kill becomes a calculation instead of an argument.
3
The cohort bounds the question
A 200-install prototype is asked about D0 behaviour, never about D1 retention. Thin data returns InsufficientData, never a pass.
The foundation
The systems you were going to build anyway, already built
Production-grade modules on one fixed template — a Manager facade, a public interface, data models. Engineers and agents write against the same contract, and extend it in-repo with the Module SDK.
Twenty backend systems, one contract
Sessions through seasons, hardened for concurrency and audited end to end. Assemble a feature from prefab modules in days, not weeks.
- Accounts
- Sessions
- Currencies
- Inventory
- Characters
- Chests
- Battles
- Rewards
- Daily rewards
- Progression
- Store & bundles
- Redeem codes
- Leaderboards
- Achievements
- Seasons
- News
- Inbox
- Presence
- Chat
- Parties
weeks → daystime from zero to a live backend
One contract, every transport
One service contract sits under the whole API. The gateway adapts it for each client, so light realtime — presence, chat, party sync — rides the same definitions.
Change the game without a redeploy
Economy values, content and flags live in runtime config. LiveOps ships a change in minutes, and every environment promotes from the one before it.
An economy that never loses a write
Currencies and economies are treated like a bank — every grant, spend, and transfer is a recorded transaction, with a full audit trail.
Studio → Game → Environment
One tenancy model for billing, members, roles and isolation. Each environment gets its own scoped database users, in managed cloud or in yours.
Supervised agents
Agents that propose. Engineers that decide.
Every agent is approval-gated. It proposes code, a flag flip, a deploy or a rollback, and a human approves before anything lands. Autonomy widens later, behind policy, once the trust is earned.
The path every change takes
- 1SpecMarkdown is the source of truth
- 2Agent implementsAgainst the Module SDK
- 3Tests prove itWritten with the change
- 4Human approvesNothing lands without it
- 5DeployWith history and rollback
Agents reach the platform over MCP tools — the same surface a human sees in the Portal. Bring your own LLM keys, and cap the model allow-list per studio.
- OrchestratoragentTakes chat from the Portal, routes to the right agent, calls MCP tools, holds session state.
- Game DesignagentGenerates ideas against a studio brief, then writes the design pack — core loop, economy, FTUE, ad creatives.
- PrototypeagentBuilds the prototype, packages the client, ships it to a store test track, and raises feedback requests.
- DevelopmentagentDesigns specs and implements features against the Module SDK, with the tests that prove them.
- DevOpsagentDeploys builds, watches cluster health, and triages incidents and telemetry.
- LiveOpsagentOperates the live game — flags, A/B tests, content and config pushes, first-line triage.
One Portal
Humans and agents work the same surface
Prototypes, development, LiveOps and DevOps in one place, scoped to Studio → Game → Environment. An uncaught exception opens its own incident — no ticket filing, no lost stack trace.
portal.vlitz.co / devops / incidents
Incidentsenv: prod
- NullReferenceException · InventoryService.Grant142 occurrences · assigned to DevOps Agentinvestigating
- TimeoutException · StoreService.Validate17 occurrences · assigned to a human developeropen
- InvalidOperationException · Configs.Import3 occurrences · fix deployedresolved
Illustration. Incidents deduplicate on exception type and location.
Prototypes
- Ideas
- Design packs
- Pipeline
- Scorecards
- Gates
- Economics
- Feedback
Development
- Features
- Issues
- Specs
- Sprints
- Agent users
- Usage & cost
LiveOps
- Feature flags
- A/B tests
- Player management
- Store & bundles
- News & inbox
- Transaction log
- Acquisition
DevOps
- Services
- Logs & traces
- Metrics
- Incidents
- Deployments
- Approvals
Take more shots on goal. Kill the bad ones early.
A small backend team, a small LiveOps team, and agents on a hardened foundation. That is the whole bet. Bring a brief and we will walk you through the funnel.