AI for builders

The workspace for agents.

Explore free AI today. We’re building toward a single workspace for models, tools and agents on Elysium.

$0Browser model fees
2On-device models
ElysiumPlanned network
roven / autoIllustrative comparison
Mode
Roven pick · examplefast
Waiting for prices…Reference comparison only; no model request is sent.
—compared with the example top model
Roven pick—
Always top model—

Example uses 500 input and 1,000 output tokens. These hosted models are not available through Roven; this is not a live router.

Illustrative estimate using public reference prices. Your request stays in this page.
Your deposit stays yours

Vault deposits are not available yet. Elysium support is planned.

No admin keys or fees · planned

Production vault contracts and their fee terms have not been established.

Your prompts stay with you

On-device chat stays in your browser. The optional hosted mode sends prompts through Roven to OpenRouter and its selected provider.

Elysium is pre-launch

Roven is building
toward Elysium.

A new home for onchain agents, designed around the speed and liquidity of the Hyperliquid ecosystem.

Explore Elysium documentation ↗
100 msTarget block time
HYPENative gas token
HyperEVMSettlement layer

Elysium is an EVM-compatible L2 being built by Kinetiq. Its team targets 300 Mgas/s, with HyperCore market-data and trading integrations planned after mainnet.

Elysium testnet · read-onlyChecking network…Connecting to the public testnet RPC.
Network data is live from testnet. Elysium wallet connections, credits and vaults are not connected. Elysium’s specifications and launch timeline may change.
How it works

One balance. Every bill your agent has.

Our planned workspace brings AI, payments and agent activity together. Financial services are not connected yet.

01

Pay for models

One place for model access. For now, use our two free on-device models without an account.

Open free chat →
02

Pay for everything else

Paying APIs and other agents from a shared balance is planned; it is not yet available.

See current limits →
03

Track every agent

Agent wallets, spending history and budgets are planned for a later release.

See the workspace →
AgentActivityDestinationCost
researcher-01Model requestAI model$0.0004
researcher-01Model requestReasoning model$0.0118
builder-07API callData service$0.0100
Illustrative example · agent billing and payments are not connected
For agents

Everything an agent needs. One workspace.

Free local AI is available now. These additional capabilities are part of the Roven roadmap.

Available

Think locally

Run SmolLM2 or Qwen in your browser without an API key.

Try the models →
Planned

Search and read

Give agents tools to discover useful information and work with it.

Planned

Read onchain data

Explore balances, transactions and contracts from an agent workflow.

Planned

Pay for APIs

Support calls to services and other agents from a common balance.

Planned

Agent wallets

Manage individual agent wallets and their spending policies.

Planned

Explore credits

Solana credits and a vault are in development and are not connected.

Read the docs →
Guardrails · planned

Limits you can understand.

Spending limits and agent controls are planned. They are not enforced by the current browser chat.

  • Daily capA planned limit across an agent’s activity.
  • Per-payment limitA planned maximum for a single payment.
  • Allowed models and payeesPlanned controls over where an agent can spend.
  • PauseA planned way to stop an agent.

Connect in one step

Available

Browser chat

Try the free on-device models without setting up an account.

Open chat →
Available

Phantom wallet

Connect your Phantom Solana account without sending a transaction.

Connect Phantom →
Planned

MCP clients

Remote agent tools and client connections are in development.

Planned

OpenAI-compatible SDK

A hosted model endpoint and API keys are not connected yet.

Models

Compare the cost. Choose what fits.

Roven’s browser models are free to use. The external catalog below is for price reference only; those hosted models are not available through Roven.

ModelInput / 1MOutput / 1MAvailability
SmolLM2 135MRuns on your device$0$0Use in Roven
Qwen2.5 0.5BRuns on your device$0$0Use in Roven
Loading external reference prices…
External reference prices are loading.OpenRouter price source ↗
For builders

Start with what works.

Explore the free browser chat, connect a Phantom Solana wallet, and inspect the public model catalog. A hosted model API and API keys are still in development.

Model catalog · public JSON
curl https://roven-psi.vercel.app/api/models
Funding · planned

Top up once. Spend on anything.

Funding with real money and spending across services are planned, not connected.

01

Wallet funding

Real-token top-ups are not available. Connect your Phantom Solana wallet to get started.

Connect Phantom →
02

Card funding

Live card payments and redeemable production credits are not connected.

Current limits →
03

Vault yield

A Solana vault is not connected and does not produce yield.

Learn more →
Vault · planned

Explore what comes next.

A Solana vault is planned. Deposits and yield are not available yet.

Read the docs →
Credits · planned

Credits for your agents.

Solana credits are not yet issued or redeemable.

Read the docs →
Questions

Briefly.

What is Roven?

Roven is an early-stage AI workspace building toward Elysium. Today it offers AI chat and a Phantom Solana wallet connection; Elysium integration is planned.

What does it cost?

On-device AI has no model API fee. Your device and network may incur their normal costs. Connecting a wallet does not send a transaction.

Do I need a wallet?

No wallet is needed for browser chat. A Phantom Solana wallet is needed for wallet connection.

Are deposits or credits real?

There is no live Solana vault or redeemable Roven credit yet.

Are my prompts stored?

On-device chat keeps conversation messages in the open tab and downloads model files from Hugging Face. The optional hosted mode sends your conversation through Roven to OpenRouter and its selected model provider.

Start here

Give your ideas room to run.