Sage AI is grounded in your live indicator state, your journal, your playbooks, and the desk’s decisions. It explains what happened, why it happened, and where in the audit log you can verify it — with citations on every claim.
Three things general AI gets wrong about trading. Three things Sage AI is built to fix.
Ask ChatGPT why your trade lost. It will invent a clean three-paragraph narrative that sounds right and references nothing. The story is fluent and unverifiable — the worst combination for a trader.
Sage AI cites a specific decision in your audit log for every claim. No grounding, no answer.
A copilot that hasn’t read your journal, your indicator state, your regime call, or your playbook is just a search engine with vibes. It cannot say anything specific to your edge because it has never seen it.
Sage AI is grounded in live state: trades, indicators, journal cohorts, regime, audit log, playbooks.
If you can’t reproduce the answer the model gave you yesterday, you can’t use it for anything serious. Without a log of what was retrieved and what was generated, the copilot is a magic 8-ball.
Every Sage AI answer logs the retrieved context, the model version, and the citations — replayable forever.
Sage AI is a constrained RAG pipeline, not a chatbot wrapper. Every answer is a function of retrieved state.
User question + active context (open chart, ranked desk row, audit event) is captured and decomposed into retrieval intents.
Vector + structured search over your trades, journal cohorts, indicator state, audit log, playbooks, and the public docs index.
Retrieved snippets are stamped with citation IDs (trade_id, audit_id, journal_id). If grounding score < threshold, model is told to refuse.
Streamed answer constrained by a grounded-only system prompt. Every numeric claim must reference a retrieved snippet.
Final answer is emitted with inline citations, then both prompt + retrievals + response are written to your audit log for replay.
Each capability has a constrained scope, a defined retrieval surface, and an answer template. If a question falls outside, Sage AI says so.
Point at any bar on the active chart. Sage AI reads the indicator state at that instant and explains what the desk would have called there.
“What did the regime matrix say at 09:31:18?”
Pulls the trade row from your journal, runs your historical setup × regime cohort, and surfaces deviations from your own playbook.
“Was this trade inside my plan or an override?”
When QuantFlow refuses or sizes down, Sage AI traces the gate that fired, retrieves the inputs, and explains in plain English why.
“Why did the vol budget gate block the entry?”
Daily and intraday briefs grounded in the regime matrix, calendar, and prior-session profile. Read in 30 seconds, replayable later.
“Brief me on the open. What changed overnight?”
Hand Sage AI an audit_id. It walks the chain backwards — signal, gates, fill, journal entry — and returns a human-readable timeline.
“Walk me through trade_4287 end to end.”
Grounded in your own playbooks (not the public internet). Surfaces the rule for the setup you’re currently looking at, with citation.
“What does the playbook say about VAL fades in vol·rising?”
Sage AI never retrieves from the public internet during an answer. The retrieval surface is fixed, scoped, and entirely under your account.
Volume Profile, QPulse, Flow Pro, ACE Squeeze, Regime Matrix, etc. — sampled at the bar Sage AI is asked about.
Every trade you’ve entered, with broker fills, R-multiple, setup tag, regime tag, and any notes you wrote.
The setup × regime heatmap your journal generates. Sage AI quotes expectancy, sample size, and recency, never anecdotes.
Every desk decision and every QuantFlow gate. Tamper-evident, replayable, with millisecond precision.
The text rules you’ve written for yourself — uploaded as markdown or built in the playbook editor. Indexed and cited.
Nexural documentation, indicator specs, glossary. Used only when the question is about the platform, not about your trades.
The shape of a tool is defined as much by its constraints as its features. Here are ours.
Sage AI never fires entries. It explains, critiques, and traces. Generating signals is the desk’s job; executing them is QuantFlow’s.
It will never tell you whether to take a trade, what size to use, or what to think about a market. That’s outside the scope it’s constrained to.
It cannot search Twitter, read news in real time, or browse external sites mid-answer. Retrieval is limited to your own data and the docs index.
If the grounding score is low or the retrieval is empty, Sage AI refuses to answer. We’d rather miss a question than fabricate one.
Your trades, journal, and playbooks are retrieval inputs only. Nothing you give Sage AI is used to fine-tune any model, ours or anyone else’s.
If something here is unclear, email us — we’ll write a better answer and ship it back to this page.
We route across a small set of frontier models depending on the capability — typically a fast model for retrieval reasoning and a larger model for synthesis. Model + version are stamped on every answer in the audit log so you can reproduce or compare.
No. Your data is retrieval input only. Nothing you give Sage AI is used to fine-tune any model — ours, OpenAI’s, Anthropic’s, or anyone else’s. We use enterprise endpoints with no-training contracts.
Sage AI refuses. The system prompt forbids generation when the retrieval score falls below threshold for the question intent. You see ‘not enough grounded context for a confident answer’ instead of a fabricated paragraph.
Yes. Every answer has a ‘sources’ tray that shows the retrieved snippets with their IDs (trade_id, audit_id, journal_id, doc_id). Click any one and you jump to the underlying record.
P50 first token is under two seconds; full grounded answer typically returns in 4–6 seconds. Retrieval and grounding stages have hard latency budgets visible in the answer trace.
Sage AI ships with the Pro and Automation tiers at a generous monthly query budget. Heavy users on Automation get unlimited queries and priority routing on the larger synthesis model.
No. The system prompt explicitly forbids it. Sage AI explains, critiques, traces, and surfaces statistics from your own journal — the decision is always yours. Anything else would put us in advice territory we’re not licensed for and don’t want to be.
Every surface compounds the others. The Swing Desk feeds the journal. The journal grounds Sage AI. Sage AI cites the indicators. The indicators feed QuantFlow.
Sage AI's first retrieval source. Every claim about your stats links back to the trades it was computed from. Your data is the moat.
The verifiable performance grounding. Sage AI cites these when it answers questions about how a regime or playbook has performed historically.
Sage AI knows what STS, composite-z, vol-shape, and the regime classifiers do because the indicator docs are part of its retrieval corpus.
Sage AI is included on the Pro and Automation tiers. Every answer is grounded in your own data, every claim is cited to a specific record, and every conversation is logged for replay.