Chapter 4
Largo
BlackOut Intelligence desk analyst wired to live GEX, flow, and positioning — structured reasoning, not generic chat.
Overview
Largo answers structured market questions using tool calls into live platform data: GEX structure, flow context, positions, regime, and Night Hawk editions.
Unlike the SPX Slayer commentary rail (push feed), Largo is pull-based: you ask, it fetches, it responds with cited tool traces. It is an analysis terminal, not a quote feed.
Route: `/terminal`. Session persists in sessionStorage so multi-step investigations survive refresh within the same browser session.
Desk layout
Largo is a single full-page chat surface — no side rails. The message log occupies the center; starter prompts appear when empty; input row sits at the bottom.
While working, LargoThinkingState replaces idle UI: rotating phrases, pipeline nodes, and live tool trace labels show which data sources are being queried.
Tool chips under each assistant message list what was actually fetched — always read them before trusting a number in the prose.
Panel reference
Every region below maps to a live UI panel on the desk. Read purpose first, then cadence, then how to consume — that order mirrors how you should scan the tool during RTH.
Message log
Center — main scroll region
Conversation thread between you and Largo with markdown-rendered answers.
What it shows
- User bubbles (your questions)
- Assistant bubbles with LargoMessageBody markdown
- Tools-used chips per answer listing invoked data sources
- aria-live region for accessibility during streaming
What you can do
- Scroll history; read tool chips to verify grounding
How to consume
Read answers top-to-bottom but verify numbers against tool chips. If get_spx_structure ran, levels should match Slayer/Thermal. If no tools ran, treat output as general commentary only. Long answers may include tables — scan headers first.
Starter prompts (Try asking)
Empty state — above input
Four fixed example questions that demonstrate good query shape.
What it shows
- SPX setup question
- Flow noise question
- Gamma map question
- Three-line structure summary
What you can do
- One-click send — populates and submits the prompt
How to consume
Use these as templates: specific, falsifiable, data-grounded. Adapt wording to your session — e.g. swap in your ticker or time window. Avoid yes/no trade permission questions.
TipGood Largo questions name the instrument, the lens (GEX, flow, regime), and what would change your mind.
Follow-up chips (Ask next)
Below last assistant message
Dynamic suggested follow-ups based on the prior answer (up to 3).
What it shows
- Contextual next questions generated with the response
What you can do
- One-click send
How to consume
Follow-ups are the fastest way to drill deeper without retyping context. Use them for stress-tests: What invalidates this? What does flow show at the wall? What changed since open?
Input row
Bottom — fixed compose area
Free-form question entry with busy state while Largo is working.
What it shows
- Placeholder marquee when idle
- Disabled/busy state during streaming
- Send button
What you can do
- Type question and submit form
- Wait for busy to clear before next send
How to consume
One question per thread turn keeps tool traces readable. If you need a multi-part investigation, use follow-ups rather than stuffing six questions into one message — routing picks tools per intent.
Tip503 responses may appear during AI budget kill-switch — retry after a minute; data desks still work.
LargoThinkingState
Replaces input area while loading
Transparency into in-flight tool calls and pipeline progress.
What it shows
- Rotating status phrases
- Pipeline node visualization
- Active tool labels (e.g. SPX desk, GEX map, HELIX flow tape, Night Hawk)
How to consume
Watch which tools activate. If you asked about flow but only get_spx_structure runs, rephrase to mention options flow explicitly. Unexpected tools may mean intent router interpreted your question broadly.
Tool trace chips (per message)
Footer of each assistant bubble
Audit trail of live data sources consulted for that answer.
What it shows
- Human-readable labels: live desk feed, SPX desk, confluence engine, GEX map, HELIX flow tape, Night Hawk, etc.
- Maps to ~70+ backend tools in LARGO_TOOL_DEFS grouped by SPX desk, flow, vol/macro, news, platform
How to consume
If chips include get_nighthawk_edition, cross-check PlaybookBoard. Missing expected tools is a signal to ask a narrower follow-up. Never treat an answer about live levels as current without a structure or GEX tool in the trace.
TipFull tool catalog lives in platform routing — UI shows a curated label subset for readability.
How it works
Each response routes through intent detection, selects tools from LARGO_TOOL_DEFS, executes live fetches, and streams synthesized prose. Session hydrate restores thread from sessionStorage on mount.
Grounded responses
Live tool calls prevent stale training-data hallucinations on levels and regime.
Streaming
queryLargoStream delivers tokens incrementally — thinking state shows progress.
Session memory
Conversations persist per browser session for multi-step analysis.
Kill-switch aware
Spend limits protect platform-wide AI budget — rare 503 during outages.
Step-by-step workflow
Use after reading Slayer header and matrix. Largo interprets; it does not replace execution discipline.
- 1
Ask specific questions
"Where is gamma flip vs spot and what changed since open?" beats "Should I buy?"
- 2
Read tool chips
Confirm get_spx_structure or get_gex ran before acting on levels.
- 3
Cross-check Slayer
Compare Largo output to active play card and walls.
- 4
Use after ambiguous flow
When HELIX and structure disagree, Largo can synthesize with get_flow_tape.
- 5
Follow up
Use Ask next chips to stress-test invalidation.
Best practices
Do
- Ask falsifiable questions about current data.
- Use for thesis stress-tests and regime explanation.
- Read tool chips on every answer.
- Prefer follow-up chips for depth.
Don't
- Don't use for entry timing alone.
- Don't treat output as trade advice or financial guidance.
- Don't assume off-hours answers use live RTH data — Largo will note staleness.
- Don't spam parallel questions while busy — wait for stream completion.
Connected tools
BlackOut is one pipeline. These desks share the same GEX, flow, and intelligence layers.
FAQ
Does Largo work off-hours?
Yes, but data may be last RTH snapshot — check tool chips and any staleness notes in the answer.
Largo vs SPX commentary rail?
Rail pushes periodic desk narrative; Largo pulls on your questions with explicit tool traces.
Open Largo
Launch the live Largo desk and apply this chapter in real time.