Open problems, and a repo to put them in.
If you want to help build this, leave an email and a line about what you'd pick up. That's the whole ask. What follows is what's being built, what help actually moves it, and what this is not.
A vault the agent can trade from and never drain.
Rienda is a self-custodial vault for LLM trading agents on Robinhood Chain. The vault holds the capital and enforces ten guardrails in contract code; the agent holds a session key that submits trade intents and can do nothing else — no withdrawals, no policy changes, no path to granting itself either. Every trade links back to the Ed25519-signed receipt of the inference that proposed it, so the decision history is checkable by someone who doesn't trust us.
Status, plainly: in development. The contracts deploy to Robinhood Chain testnet (chain id 46630) before anything else, and mainnet waits on an external audit and a legal review — however long those take. Nothing accepts deposits today and nothing has traded, so there is no track record to show you. The signed-inference rail underneath has been on Solana and Base mainnet since 2026-05-21; that part is real and has callers.
This is not a job posting. VDM Nexus is early and pre-revenue, there is no funded headcount, and no compensation, equity, tokens, or airdrop is on offer — contributing does not earn any of them and nothing on this page should be read as implying otherwise. Nobody is being hired, contracted, or promised future consideration.
What there is: unsolved problems that are written down, code review on what you send, and a commit history with your name in it. The published packages are MIT and stay that way. If the funding situation ever changes, that gets written down publicly before anyone is asked to act on it.
Four places, specifically.
Not a list of every role a company might one day have. These are the four where an outside contribution changes what ships next. If your thing isn't here and you still think it matters, say so in the box below — a good argument beats a category.
The vault and the policy engine
M1 is the vault contract plus the policy engine that runs ten guardrail checks on every trade intent — position caps, a daily realized-loss limit, drawdown throttles, an oracle price band. Separately there is Uniswap v4 hook work behind the fee-burn pool. Foundry tests are the currency: a failing test showing a guardrail can be walked around is worth more here than a new feature.
Agent host, model router, x402
The agent is the untrusted party by design, so the interesting work sits at its boundary — session-key handling, intent construction, what happens on a revert, how a retry avoids becoming a second order. The model router and the x402 payment path are both live code you can read today: eight MIT packages on npm and PyPI.
Backtesting and execution quality
The compute budget meters an agent's inference allowance against its realized PnL, which only means anything if the PnL is measured honestly — slippage, fees, fill assumptions, survivorship. The design already assumes most strategies lose money. Finding out exactly how, in a backtest, before any capital moves, is the useful contribution.
Try to break it
No third-party audit is scheduled. The vault contracts are private until the testnet deploy, so today this means the published rail: receipt signing, the five-check verifier, the paywall middleware, the facilitator path. When the contract repo opens, guardrail bypasses and session-key escalation are the first things worth attacking. Reports go to security@vdmnexus.com.
Email, and one line.
The line is the part worth writing. A repo, a handle, or the specific piece you'd take — that's what makes a reply possible instead of a thank-you note.
The code and the chat are both open.
An issue or a pull request is a fine first message. So is a question in the Telegram.