Trading agents

Research that knows your book.

Agents are bounded research tools, not autonomous traders. They read what they are told to read, write to the analyst log, and pass through the same approval and risk machinery a human order does.

§ 01

What an agent does

An agent assembles a take on a symbol using a configurable set of analyst tools, returns a signal with a confidence value, and writes the full reasoning to an analyst record.

Market analyst -- price action, volume, technicals, AiQ commodity context.
News analyst -- recent headlines, source-linked, with sentiment attribution.
Social analyst -- Reddit/forum reads, scoped to relevant communities.
Fundamentals analyst -- balance sheet, earnings, key ratios.
Risk manager -- final consolidation; can downgrade a signal it does not trust.

§ 02

What an agent does not do

Agents are tools, not authority. The same machinery that gates a human order gates an agent order.

Never auto-executes without an explicit per-agent opt-in and a broker connection.
Never bypasses risk gates, position limits, or paper/live toggles.
Never holds opinions without traceable sources -- every signal is auditable.
Never accesses credentials, account secrets, or private data outside the platform.

§ 03

Operating discipline

Agents inherit the platform's operational posture. Lifecycle, persistence, and limits all live in the audit log.

Agent state persists across restarts via the engine restore path.
Concurrent analyses run thread-safely with per-call config isolation.
Daily-loss limits and max-concurrent-position caps apply per agent.
Failure modes are explicit: rejected, skipped, manual-approval-required.

Next step

Try an agent against your watchlist.

Start with paper-mode agents on the symbols you actually trade. Approve, reject, and tune confidence thresholds before any live execution.

Plan agent access