How Loxley AI auto-buy works on Robinhood Chain: pair detection, quoting across V2 pairs and Uniswap V4 hook pools, layered safety scanning, transaction simulation, broadcast, confirmation and notification — stage by stage.
The auto-buy module is the heart of the Loxley AI Telegram trading bot. This document describes the full execution flow for a member's automated buy on Robinhood Chain. The same conceptual pipeline drives manual buys, with different gating rules (see the section on manual vs automated at the end).
New liquidity appears on Robinhood Chain in two forms: classic V2-style pairs created through the factory contract, and Uniswap V4 pools initialized on the pool manager. Loxley indexes pair-creation events and, crucially, anchors each token's identity to its creation transaction. That anchor matters later: it lets the analyzer reconstruct who deployed the token, who funded the liquidity, and which wallets sniped within the first blocks — before events can be obscured.
Not every new pair becomes a candidate. The pipeline drops tokens whose deployment bytecode matches known-scam family signatures, tokens blacklisted by operator or platform policy, and pairs whose liquidity is below the operator-configured minimum. Candidates pass to quoting immediately; nothing waits for a cron tick.
Quoting is hook-aware. For V2-style pairs the quote is a standard constant-product estimate. For V4 pools the quoter accounts for the pool's hook contract, because hooks can modify swap amounts, take fees in either direction, or enforce custom logic around beforeSwap/afterSwap. A naive quoter that ignores hooks systematically misprices these pools — one of the most common failure modes of multi-chain bots grafted onto Robinhood Chain.
Before any automated buy, the safety scanner assembles a layered verdict from live on-chain reads:
If any layer is unreachable — RPC flaky, scanner timeout — the verdict is degraded, and degraded verdicts are never cached or reused. Automation treats degraded as a stop, not as "probably fine".
Approved buys are signed for the member's embedded wallet, broadcast through redundant RPC endpoints with failover, and confirmed on-chain. The member receives a Telegram notification with entry details; the position is registered immediately in the tracking layer so the sell engine watches it from the first block of ownership.
When an operator enables an AI autonomous strategy mode for a channel, lane-level shadowing prevents double execution: the strategy wave becomes the single purchasing voice for that lane and plain auto-buy stays silent (no duplicate buys, no duplicate member messages). Kill switches let operators halt auto-buy per channel or globally without redeploying.
Loxley deliberately separates the two. Automated unattended buys are hard-blocked by the safety gate. Manual buys — where a human taps the button — are never hard-blocked: the interface shows explicit warnings on dangerous or critical ratings, and the user decides. This keeps informed risk-taking possible while stopping asleep-at-the-wheel losses.
Detection is event-driven. Quoting and scanning are parallelized. Simulation is the deliberate bottleneck — it exists to be slow enough to be right. We do not quote a single number of milliseconds because chain conditions govern reality; we do promise the pipeline has no idle sleeps, no polling gaps, and no silent skips.
Next: the sell engine → All docs
Built for operators, not slippage goblins. Talk to us →