Talk to Injective
A friendly AI chat that explains Injective in plain English using live mainnet data — wallets, markets, governance, staking. Just ask, and it reads the chain for you.
Videos

Tech Stack
Description
Talk to Injective is a read-only AI chat that explains the Injective blockchain in plain English — using live mainnet on-chain data. No trading, no signing, no private keys. You just ask, and it reads the chain and explains what's going on.
The problem. Injective is a trading-first chain, so almost every tool assumes you already speak the language — funding rates, perps, subaccounts, governance params, staking APR. For a newcomer that's a wall. This is the opposite: a chat that meets you where you are and defines terms inline as it goes.
What it does. Ask about any wallet, market, governance proposal, or staking — in normal language. The AI agent picks the right on-chain read, fetches live data, and explains the result for someone new to crypto. It never invents numbers: every figure comes from a real tool call against mainnet.
Four live tools:
- Portfolio — token balances + open derivative positions for any inj1 address
- Market snapshot — mark price, funding rate, open interest for any perp market
- Governance — active proposals, or any single proposal explained plainly
- Staking — estimated INJ staking APR + total bonded, plus a wallet's delegations and pending rewards
Safety. Strictly read-only. When you ask where to actually stake, trade, or vote, it points you to the official Injective apps (Hub and Helix) — it never acts for you. Optional Keplr/Leap connect is address-only; the app never requests a signature.
Progress During Hackathon
Built end-to-end during the sprint:
- Designed and shipped the chat agent with 4 read-only Injective tools (portfolio, markets, governance, staking)
- Integrated @injectivelabs/sdk-ts against mainnet across the Bank, Exchange/derivatives indexer, Governance, Staking, Mint and Distribution modules
- Wired the agent with the Vercel AI SDK + OpenRouter (Gemini 3 Flash) using tool-calling, so the model only speaks from live data
- Normalized raw chain values (base units, denoms, chain-scaled prices) into human-readable numbers to keep answers accurate
- Built the terminal-style chat UI: streaming answers, quick-action prompts, clickable links, inline term definitions for newcomers
- Added optional read-only Keplr/Leap wallet connect to point prompts at your own wallet
- Wrote a test suite (29/29) covering normalization/resolver logic plus live mainnet integration reads
- Deployed live on Vercel
Fundraising Status
Not raising. Solo, self-funded hackathon project built for the Injective Solo AI Builder Sprint. Open-source under the MIT license.