AskAura

A conversational research assistant grounded in your portfolio, your corpus, 13F flows, SEC filings, and live X/web data — now multi-turn, multimodal, and chart-aware. Available everywhere via the floating launcher.

AskAura is a research assistant that uses your own data first — your portfolio, your curated Substack/X/YouTube corpus, 13F flows, and SEC filings — then reaches for the live web only when needed. Every answer cites its sources inline.

It’s a full conversation, not a one-shot search box: ask a question, then keep digging with follow-ups, attach a chart or a filing, and get mini price charts for the names it discusses.

There are two ways to use it:

• Floating launcher (✦ AskAura pill, bottom-right of every page). Opens a side panel on top of whatever you were doing — great for quick lookups without losing your place.
• Dedicated page via the AskAura button in the top nav (it has its own accented tab — no menu digging). Best for long deep-dive sessions, with a full history sidebar and collections.

Both share the same backend, history, and budget.

How it works

You type a question. The agent decides which tools to call, gathers data, then synthesizes a cited answer — streaming the tool calls and the answer as it goes.

There are two modes:

Mode	Model	Best for	Typical cost
Default	Gemini 2.5 Pro	One-shot questions, quick lookups, fact checks	$0.01–0.10
Deep research	Claude Sonnet 4.6	Multi-step analysis, theses, bull/bear cases, filings	$0.15–0.65

Toggle deep mode with the checkbox below the question box. Some questions auto-upgrade to deep mode, and attaching a PDF always uses deep mode (see below).

Conversations

AskAura holds a thread. After an answer lands, just keep typing — your follow-up carries the prior turns as context, so “now compare that to NVDA” or “what’s the bear case?” just works.

Follow-ups

The thread and composer stay put after each answer. Ask as many follow-ups as you like; the last several turns travel with each new question.

Stop

While an answer is streaming, the Ask button becomes Stop ⏹. Click it to halt — it stops the work (and the billing), not just the display.

Edit & regenerate

Hover any past question and click it to edit and re-run from that point. Resubmit it unchanged and you’ve effectively regenerated the answer.

New thread

+ New question (history sidebar) starts a fresh conversation. Clicking a saved session reloads it — and you can continue it.

Attachments

Bring your own context. Attach files three ways: the 📎 button, drag-and-drop onto the composer, or paste an image straight in (Ctrl/Cmd-V a chart screenshot — the fastest path for technical analysis).

Images

Charts, screenshots, tables. Up to 4 images, ≤5MB each. “What’s this pattern?”

PDF

A 10-K, earnings deck, or research note (≤15MB). Switches to deep mode automatically for long-document reasoning.

CSV / TXT

Holdings, data tables, notes (≤200KB). Read as text and folded into the question.

A PDF and images are mutually exclusive (one PDF or up to four images per question), plus an optional small CSV/TXT. Attachments show as chips in the composer before you send and as 📎 name on the question afterward. (Saved history keeps the filename, not the file.)

Ticker autocomplete

Type @ or $ followed by a few letters in the question box to get a ticker typeahead from your universe — arrow keys to move, Enter/Tab to accept. It inserts the clean symbol so your question references exactly the name you mean.

Tools available

The agent has 12 tools to choose from per question:

Tool	Source	When it’s called
get_quote	Yahoo Finance	”What’s X trading at”
get_price_history	Yahoo Finance	”How has X performed over Y”
render_chart	Yahoo / Polygon	”Show me a chart of X”, “chart A vs B” — draws a real chart inline
get_fundamentals	Yahoo Finance	Revenue, EPS, margins (last 8 quarters)
get_filings	SEC EDGAR	10-K, 10-Q, 8-K with direct links
search_research_corpus	Your Substack/YouTube/X library	”What do writers think about X”
get_writer_calls	The Calls ledger	”Who’s long/short X”, positioning + each writer’s track record
get_thirteenf	The 13F fund universe	”Who’s buying X”
get_portfolio_context	Your holdings	”How am I positioned in X”
get_market_snapshot	Macro engine	Cross-asset regime, sectors, divergences
web_search	Google-grounded (via Gemini)	Static facts, editorial articles
x_search	Grok / xAI live search	Real-time X/finTwit sentiment, breaking news

Each tool call appears live in the answer panel as it fires. Once the answer is in, the tool log collapses into a one-line summary (Used 7 tools · 25 sources) — click to expand.

Auto-upgrade to deep mode

Thesis-style questions benefit from the slower-but-smarter Claude orchestrator, so the page flips deep mode on automatically when the question contains things like:

• deep dive, thesis, bull case / bear case
• should I buy / sell / hold / own / short / trim / add
• compare X vs / and / to Y
• valuation analysis, do a full / complete / comprehensive analysis

…and whenever you attach a PDF. An orange banner tells you when it happened — uncheck the Deep research toggle for the next question to override.

Citations & sources

Every quantitative claim has a [N] chip linking to its source. The agent is instructed not to invent numbers — anything it states should be backed by a tool result.

• Hover a [N] chip for a quick preview card (favicon, domain, full title).
• Card row at the top of the answer shows the favicons of the unique domains consulted (yahoo, sec.gov, x.com, …).
• Sources panel at the bottom lists every source, [N] 🌐 domain · title, with “Show all N →” to expand.

When an answer leans on your own corpus or the Calls ledger, a grounding line appears — e.g. “Grounded in 15 takes · 9 writers · last 11 days” — so you can see how much real, human-sourced research is behind it. The footer shows how long the answer took.

Suggested follow-ups

After each answer, three contextual follow-up chips appear — specific to what you just asked (e.g. “What’s the bear case?”, “Who’s short this?”, “How does it compare to peers?”). Click one to keep the conversation going.

The composer also shows a few starter chips when the thread is empty, to kickstart common patterns.

Charts

AskAura renders charts two ways — one automatic, one on request.

On-demand charts

Just ask to see one: “show me a chart of NVDA,” “chart AAPL vs MSFT over 1Y,” or “plot the 10Y yield.” The agent draws a real line chart inline, right below the answer:

Single name

A price chart over your chosen window, shaded green/red by direction.

Comparison

Pass more than one ticker and they overlay as a rebased % comparison (everything starts at 0%), with a per-name legend showing each one’s return.

Macro too

Not just stocks — “plot the 10Y yield”, oil, gold, the dollar index, or bitcoin all work.

Interactive

Hover for a date + value tooltip, switch the window with the 1M / 3M / 6M / YTD / 1Y / 3Y toggles, and click a name to open its entity page.

This is for when you want to see the picture. If you just want the numbers or performance facts, a normal “how has X done?” question returns those as text instead.

Charts are saved with the conversation — reopen a past session and they re-render with fresh data.

Auto “Mentioned” strip

Separately, when an answer discusses specific tickers, a compact strip of mini sparklines (last ~3 months, with % change) appears beneath it for the names it actually references — no need to ask. Click one to jump to that name’s entity page. Only genuine tickers are charted; passing mentions of all-caps words are filtered out.

Export & share

Under every finished answer:

• ⧉ Copy — plain text (question + answer with [n] citations + numbered sources)
• ⧉ Copy as Markdown — same, formatting preserved, sources as links
• ⎙ PDF — a clean, light-themed print view (question, formatted answer, sources) → Save as PDF

History & collections

Every Q&A auto-saves to your per-user history (signed-in users; anonymous questions aren’t persisted). The left sidebar shows:

• ★ Pinned (coming soon — backend supports it, UI pending)
• 📁 Collections — group related sessions into named spaces (see below)
• Today / Yesterday / Last 7 days / Older for everything else, most-recently-updated first

Collections / Spaces. Hover a history row and click the 📁 button to move that session into a named collection (or clear it). Collections show as their own groups at the top of the sidebar — handy for organizing a research thread by theme, ticker, or project.

• Search the history with the input at the top (matches title, question, ticker).
• Click a row to reload the full conversation (all turns, sources, tool log, cost) and continue it.
• Hover a row for 📁 move and × delete buttons.
• Collapse the sidebar with the ◧ button for more room.

Cost guardrails

• Per-request hard cap: $0.50 default,$1.50 in deep mode. If the running cost would exceed it, the agent stops and returns what it has.
• Time budget: deep research finishes within a bounded window and emits a complete answer — or a clearly-labeled partial with a “cut short — ask a narrower follow-up” note — rather than ever hanging.
• Cost line under every answer: claude · $0.282 · 19,876 tokens · answered in 1.4min.

What it’s good at

• Multi-source synthesis — “deep dive on $X” pulls quote + fundamentals + filings + corpus + 13F + X chatter + web news in one shot, with a structured bull/bear summary.
• Document analysis — drop a 10-K or paste a chart and ask about it directly.
• Real-time sentiment — “what is finTwit saying about NVDA right now” hits the Grok-powered x_search tool for live X data.
• Personalized framing — ask about a name you hold and the agent weaves in your position size and P&L.

What it’s not good at

• Live order-book questions — no L2 data or market-making signal.
• Backtests — it retrieves and reasons over data; it doesn’t run simulations.
• Predictions — it summarizes what consensus thinks; it won’t call where a stock is going.

Behind the scenes (for the curious)

• Default mode runs Gemini 2.5 Pro with function-calling; deep mode runs Claude Sonnet 4.6 with tool_use chains (up to 8 gathering turns, with a reserved final turn that always synthesizes from what it gathered).
• Follow-ups thread the last several turns as conversation context; tools re-run fresh each turn.
• Attachments ride on the question as native image/PDF blocks (Claude) or byte parts (Gemini); CSV/TXT is folded into the prompt.
• x_search calls xAI’s Responses API with the Agent Tools framework.
• Streaming uses Server-Sent Events — tool calls/results stream as they happen, the answer streams token-by-token, and markdown renders on completion. Stop signals the agent to halt server-side, not just client-side.