Documentation

Build with Roven.

Explore the models, APIs, and wallet tools available in Roven.

Quick start

  1. Open AI chat. No wallet or personal API key is needed.
  2. Choose an on-device model or a shared free hosted model.
  3. For onchain experiments, open the wallet page in a browser with Phantom.
On-device models run on your own device. Downloads need an internet connection, device storage and sufficient memory. Hosted models use OpenRouter.

Free models

SmolLM2 135M is the smaller option, with an approximately 180 MB first download. Qwen2.5 0.5B downloads approximately 790 MB and needs more resources. Both run locally; prompts are not sent to a model provider.

Availability depends on your browser and device. A small model can make mistakes, so verify important answers.

Free models with OpenRouter

Roven chat sends hosted requests through Roven to OpenRouter. Choose Free hosted · Auto for OpenRouter’s openrouter/free router, or choose a specific free model from the dropdown. Auto selects an available free model; it does not guarantee a particular model or provider.

Hosted chat uses a shared connection managed by Roven; visitors do not need an OpenRouter key. When you send a hosted request, your conversation passes through Roven to OpenRouter and the selected model provider. Their data policies and free-tier rate limits apply. Shared capacity is limited and may be rate-limited.

Open Roven chat →   How the free router works ↗

Currently listed free text models

Loading OpenRouter’s public catalog…

    These entries have a :free model ID, text input and output, and zero listed input and output price. Catalog data can be cached for a few minutes. OpenRouter controls availability; being listed does not guarantee a successful response.

    curl https://rovenagent.xyz/api/free-models

    Read the model catalog

    The public catalog and health endpoints return JSON. They do not generate model responses.

    curl https://rovenagent.xyz/api/models
curl https://rovenagent.xyz/api/health

    External price reference

    The public reference-price endpoint reads model prices from OpenRouter. Values are shown in USD per million tokens and are cached briefly. These are external reference prices, not prices for hosted inference through Roven.

    curl https://rovenagent.xyz/api/reference-prices

    If the upstream catalog is unavailable, the endpoint returns HTTP 503. No prompt or conversation is sent to the price provider.

    Building toward Elysium

    Elysium is a pre-launch EVM-compatible L2 from Kinetiq, designed to settle to HyperEVM and use HYPE for gas. Its team targets 100 ms blocks and 300 Mgas/s. HyperCore market-data and trading features are planned after mainnet.

    Roven reads the latest testnet block using Elysium’s public JSON-RPC. The endpoint verifies the chain ID before returning the block number:

    curl https://rovenagent.xyz/api/elysium-status

    The response includes chain ID 99801, the latest block as a decimal string, and the time it was checked. Data may be cached briefly. If the RPC is unavailable or reports a different chain, the endpoint returns HTTP 503.

    For direct integration: Elysium testnet’s public RPC is https://testnet-rpc.elysium.kinetiq.xyz, chain ID 99801, and the testnet gas token is HYPE. Mainnet network details remain unpublished; use a provider endpoint for production traffic.

    The Phantom wallet connection uses Solana.

    Elysium documentation ↗

    Solana wallet

    Connect Phantom’s Solana account to share its public address with this page. Connecting does not sign a message, send a transaction or create an authenticated Roven account.

    Connect Phantom →