Safety, privacy, and limits
FlyTradr treats model output as a proposal, not as product authority. Consequential decisions remain behind deterministic validation, account permissions, current-state checks, and explicit user actions.
Bounded workspace context
The assistant receives context approved for the active workspace. Examples include supported configuration fields, selected testing summaries, and approved product knowledge.
It does not receive a generic database, browser, broker, or HTTP tool.
Redaction and data minimization
- Provider messages pass through shared redaction.
- Context builders select only fields required for the active capability.
- Operational AI audits store metadata such as capability, model, status, latency, token use, and error category.
- Complete raw prompts and raw model responses are not retained in operational audits by default.
- Controlled sanitized diagnostic capture is off by default and, if enabled for an investigation, has a maximum seven-day retention period.
Do not enter broker credentials, API keys, passwords, authentication tokens, or other secrets into the assistant.
Validation and confirmation
Structured responses are parsed and checked against strict schemas. Strategy drafts and proposals then pass domain validation before they can be shown as supported.
The assistant cannot use prompt text to bypass workspace permissions, tier entitlement, ownership checks, strategy validation, proposal expiry, or explicit confirmation.
Financial boundaries
FlyTradr AI does not provide investment advice, trading signals, market predictions, profit guarantees, a universally best setting, autonomous broker orders, or withdrawals.
Testing explanations are educational. Historical, simulated, and paper-trading results do not guarantee live performance.
Service controls
FlyTradr can disable an individual AI capability pack or activate a platform-wide AI kill switch without changing your normal platform data.
For the wider public security posture, see Security and Reliability.