Trading Infra

Signal firewall for trading agents.

SignalSieve checks market posts, token metadata, alerts, news, and quote feeds before an AI agent is allowed to trade on them.

Analyze signal View proof

Firewall status Checking analysis engine...

input market/social/feed

scan model + rules

output allow / warn / block

safeSignal { "action": "block", "riskScore": 94 }

fake listing claim prompt injection stale news replay bot sentiment wave feed disagreement thin liquidity trap fake listing claim prompt injection stale news replay bot sentiment wave feed disagreement thin liquidity trap

System

One check before the agent sees the signal.

poisoned trading inputs in the local dataset

agent actions: allow, warn, or block

risk-score ceiling before a signal is rejected

Ingest

Market inputs arrive with source type, asset, text, timestamp, and context.

Sieve

The firewall inspects manipulation patterns and market stress before passing signals downstream.

Gate

The downstream agent receives only the safe signal and recommended action.

Input surface

Trading inputs are attack surfaces.

A trading agent does not only read prices. It reads claims, metadata, scraped copy, social clusters, and third-party APIs. SignalSieve treats every one of those streams as untrusted until it passes the firewall.

Live analyzer

Paste any trading signal.

input stream source

Asset pair Source type

Source name Signal text Attach live Bitget spot ticker when the asset pair is available

Attack sample

asset expected

Loading...

firewall output No run yet

IDLE

Risk score 0 / 100

Paste a signal or choose a sample, then run the firewall. The output is what a trading agent would receive before acting.

Proof of run

Human verdict. Agent payload.

Each analysis writes an audit entry and returns a constrained safeSignal with confidence, poisoning types, and action limits.

evidence 0 findings

safeSignal JSON

{}

Audit