Bellitus

Creator
@official
Type
Oracle
Sends live market context to an LLM each cycle and proposes an allocation, but never trusts it directly -- a code-enforced policy gate re-derives every constraint before anything reaches a trade.
Oracle
Send an LLM the bot's candidate universe and current holdings, let it propose an allocation with a reason for each pick, then run that proposal through a code-only policy gate that never trusts the model directly — only a proposal that survives every rule reaches a trade.
Candidate tickers, tag scores, current holdings.
Proposal
MSFT 28% · VLO 22% · JPM 18%
“Top-ranked for momentum and trend, balanced with a quality name for stability.”
The model proposes tickers, weights, and a rationale.
Code re-checks every rule before anything trades.
Gate result
Accepted
Rejected or malformed proposals hold — never a fabricated fallback trade.
Rejected or unclear proposals hold the current position.
Oracle sends an LLM its candidate universe, tag scores, and current holdings each cycle and asks it to propose an allocation with a rationale — then re-validates that proposal in code before anything ever trades. The following slides walk through a real, gate-accepted example. Oracle is currently available as an official bot only, not yet open to custom user-built configurations.