Open source / Terminal-first / Bring your own model
ROVANTA: AI research agent for Robinhood Chain & Ethereum.
Run locally from your terminal. Ask a question and ROVANTA calls read-only research tools, gathers market and on-chain data, and returns a report where every claim is sourced.
ROVANTA token / $ROVANTA / Paired with ETH
0x518867Dc98541547ebc5383f00e954d837967e18Always verify the contract address. Not investment advice.
$ git clone https://github.com/Wangdev0/rovanta && cd rovanta$ npm install && npm run build && npm link$ rovanta config init$ rovanta research "What is the liquidity profile of ETH?"$ rovanta research "What is the liquidity profile of ETH?"
- ✓ search_tokens188ms
- ✓ get_market_data412ms
- ✓ get_liquidity655ms
- ✓ get_volume297ms
- ✓ generate_research_report1.9s
## ETH liquidity profile
- FACTDeep DEX liquidity across major pools [E3]
- FACT24h volume retrieved from CoinGecko [E2]
- INFERENCELiquidity is concentrated in a few large pools
- UNKNOWNInsufficient data on off-chain order books
Sources: E1–E4 · coingecko, dexscreener
$
01 / Vision
Markets move faster than human research.
- 01Thousands of protocols.
- 02Millions of transactions.
- 03Constantly changing liquidity.
- 04New narratives every day.
Humans cannot manually investigate everything.
ROVANTA gives AI models the tools to investigate it.
02 / How it works
Your model reasons. ROVANTA provides the instruments.
ROVANTA is not a model. It is the layer between a model and the market: tools, data, context and a research loop that turns a question into evidence.
YOUR MODEL
OpenAI-compatible, Anthropic or Google. Provides reasoning.
ROVANTA
Orchestration, context, research workflows, structured output.
RESEARCH TOOLS
Typed, validated tools the model can call.
MARKET DATA
Chain RPC and public data providers via adapters.
INSIGHT
A report where every claim is labelled and cited.
Six steps
- 1Install the CLI and run rovanta config init.
- 2Connect your own AI provider.
- 3Run rovanta research with a question.
- 4ROVANTA gathers structured information.
- 5The model analyzes the evidence.
- 6Read a sourced report in your terminal.
Research loop
QUESTIONPLANTOOLSEVIDENCEANALYSISREPORT
The model only reports what the tools retrieved. Everything else is labelled inference or unknown.
03 / Research tools
Fourteen tools. One contract.
Typed input and output schemas. Validated on every call. Extensible registry.
get_token_metadata
Name, symbol, decimals, chain, contract and supply for a token.
get_token_price
Current USD price and 24h change, with provider timestamp.
get_market_data
Market cap, FDV, volume, supply and price change snapshot.
get_volume
Latest 24h volume plus daily volume history.
get_liquidity
Total DEX liquidity and the pools behind it.
get_transactions
A single transaction by hash, or recent ones for an address.
get_wallet_activity
Read-only balances and recent activity for a public wallet.
get_protocol_metadata
Category, chains, audits and TVL history for a protocol.
get_contract_info
Bytecode, verification status and token details for an address.
get_chain_status
RPC reachability, latest block, gas price and latency.
search_protocols
Find DeFi protocols by name or keyword.
search_tokens
Resolve a name, symbol or address into candidate tokens.
compare_assets
The same market fields for 2 to 5 tokens, side by side.
generate_research_report
Compile gathered evidence into a labelled report.
export const getLiquidityTool = defineTool({
name: "get_liquidity",
description: "Get on-chain DEX liquidity for one token...",
category: "liquidity",
input: z.object({
chain: z.string().optional(),
address: z.string().optional(),
symbol: z.string().optional(),
}),
output: LiquidityDataSchema,
async run(input, ctx) {
return ctx.data.liquidity.getLiquidity(toTokenRef(input));
},
});04 / Evidence
Evidence, not guesses.
Every claim in a ROVANTA report carries a label. You can see what was retrieved, what was reasoned, and what is still missing.
Directly supported by data a tool retrieved, with a citation to that evidence.
“Data shows...”
A reasoned interpretation of the facts. Stated with hedged language, never as certainty.
“This may indicate...”
Not answerable from the evidence gathered. Said plainly instead of guessed.
“Insufficient data to determine...”
05 / Local-first & security
Local-first by design.
The safest secret is one that never leaves your machine.
Your key stays on your machine
Provider keys live in ~/.config/rovanta/config.json (mode 0600) or in environment variables, and go straight to the provider you chose. There is no ROVANTA server to receive them.
Read-only wallet lookup
Run rovanta wallet <address> to inspect balances and activity. No private keys, no signatures, no transactions.
Tool output is untrusted data
Anything retrieved from the outside world is passed to the model as data, never as instructions.
No arbitrary code execution
The model can only call registered, schema-validated tools. It cannot run code, shell commands or scripts.
- SYSTEM INSTRUCTIONSFixed rules set by ROVANTA
- USER REQUESTYour question
- TOOL DATAUntrusted. Quoted, never obeyed
- MODEL RESPONSEReasoning over the above
Each channel is kept separate, so text inside a token description or web page cannot rewrite the model’s instructions.
06 / Ecosystem
Initial ecosystem support: Robinhood Chain.
Chain access runs through a configurable RPC and a small ChainAdapter interface, so adding another EVM chain is a configuration and adapter change, not a rewrite. Support is technical integration only and implies no partnership or endorsement.
Data adapters
Public data providers plug in behind typed adapters. Swap any provider without touching the tools.
DefiLlama
Protocol and TVL data
DexScreener
DEX pools and liquidity
CoinGecko
Token prices and market data
07 / Where this goes
From research to infrastructure.
Today, ROVANTA helps AI investigate markets. Tomorrow, intelligent systems may continuously research, evaluate and interact with financial infrastructure. We are building toward that layer.
Only the first stage exists. The others describe intent, not shipped features.
- 01Today
Research
AI models investigate markets on request, with labelled evidence.
- 02Direction
Persistent research agents
Event-driven research that keeps running, with structured market memory.
- 03Direction
Autonomous workflows
Agent-to-agent research, market intelligence and portfolio intelligence.
- 04Direction
Execution infrastructure
Audited, permissioned rails for intelligent systems. Disabled today.
08 / Open source
Open source. Runs on your machine.
The research engine and the rovanta CLI live in a public GitHub repository under the MIT license. Clone it, build it, and research from your own terminal.
MIT licensed
The engine and the CLI are open source. Read it, fork it, run it, ship it.
Self-hosted by default
Install from source and run it on your own machine. No account, no hosted service, no key vault.
Bring your own model
OpenAI-compatible, Anthropic or Google. Use the provider and model you trust.
Contribute
Add a research tool, a data provider or a chain adapter through the typed registries.
$ rovanta research "<question>"$ rovanta wallet <address>$ rovanta chain status$ rovanta tools$ rovanta providers
ROVANTA