Documentation
The Assistant
A chat pane beside the dashboard on every tab. Connected to a language model — the hosted app’s free budget, your own cloud API key, or a model on your own machine — it answers from your holdings and what is on your screen, and can drive the dashboard for you.
Connecting a provider
On the hosted app you do not have to: a new session answers straight away on gpt-5.6-luna, paid from a small free budget on our own key, with a donut beside the chat showing how much is left (what that means for your data). Connect your own provider when you want a different model, a bigger budget, or inference kept on your machine — and on the desktop app, which has no free budget, from the start.
Open ⚙ Settings → Assistant — on the hosted app, first switch Model source from Built-in free to Your own provider — and pick one of five providers:
| Provider | Setup |
|---|---|
| OpenAI | Paste an API key. Offers gpt-5.6-luna (the default), gpt-5.6-terra, gpt-5.6-sol and gpt-5.5. |
| Anthropic | Paste an API key. The main slot offers claude-opus-5 (the default), claude-fable-5-1, claude-fable-5 and claude-sonnet-5; the fast slot claude-haiku-4-5 or Sonnet. Prompt caching is built in, so follow-up questions are billed at the provider's cached rate. |
| OpenRouter | Paste an API key; any model in its catalogue. The model field suggests the catalogue once a key is saved. |
| Ollama | Local inference on the same computer as the desktop app, with no per-token charge. A running server is detected on localhost (URL configurable); installed models appear in a dropdown, and on desktop new ones can be downloaded from the dialog from a curated list with disk sizes. |
| LM Studio | Also local and detected automatically — loaded models appear in the dropdown. |
Local-server detection is live: while the Assistant tab is open with Ollama or LM Studio picked, a model you load appears in the dropdown within seconds. The hosted app calls the model from our server and cannot reach localhost on your machine; a local model there needs a tunnel — see Local models with the cloud app.
API keys you enter are not mirrored to your browser: the hosted app stores them encrypted for your session or account, and desktop stores them encrypted in its local database (see Privacy & your data). When Fortunest desktop and Ollama or LM Studio run on the same computer, model inference stays there. A hosted Fortunest session still stores portfolio data on the hosted service, and reaching your model through a tunnel sends context through that tunnel. Cloud AI, tax calculations, sharing and hosted sync remain separate outbound services.
Press Save and the assistant is ready. The status dot beside the chat pane's AI Assistant title is green when it can answer and red when it cannot — hover it for the reason. Nothing calls a model and no tokens are spent while it is red. The word context beside the header controls shows how full the model's window is; see What the assistant sees.
The free budget, and what it buys
There are three sizes, each double the one below it, and only Pro's refills. The budget is metered in what the calls actually cost, and cached prompt tokens — the unchanging part of the context on a follow-up question — are charged at a tenth of the normal rate, so a conversation goes much further than its first question suggests.
| Who | Budget | Refills? |
|---|---|---|
| Guest — no account | The base amount, once per browser session | Never |
| Free account, address not confirmed yet | The guest amount | Never. Opening the confirmation link emailed at sign-up moves you to the row below. |
| Account on its 14-day Pro trial | The Pro amount, once | It does not refill: one allowance for the Pro trial, available once your email address is confirmed (the 14 days count from sign-up). Accounts created with Continue with Google or Continue with Apple start here — the address is already confirmed. When the trial ends the free account budget in the row below applies again — what the trial spent does not count against it, and an account that had already used some of that budget before its trial goes back to what was left. Getting Pro during the trial starts Pro’s first monthly allowance from zero. |
| Free account | Double the guest amount | Granted once, when you confirm your email address. It is a fresh budget, not a top-up: what you spent as a guest does not follow you. |
| Pro | Double again — four times the guest amount | Every month, on the day of the month you subscribed; ⚙ Settings → Assistant shows the next date under the bar. |
How many questions that is depends on the model, because the models differ in price by more than twenty-five times. These estimates assume conversations of about ten questions and answers of roughly five hundred words:
| Model | Guest | Free account | Pro, per month |
|---|---|---|---|
gpt-5.6-luna — the default | ~100 | ~200 | ~400 |
gpt-5.6-terra | ~10 | ~20 | ~40 |
gpt-5.6-sol — the most expensive | ~4 | ~8 | ~16 |
GPT-5.6 Sol is on an OpenAI promotion until at least 21 November 2026 — $4 per million input tokens and $20 per million output tokens instead of $5 and $30 — and the budget is charged the promotional price while it lasts, so Sol goes roughly 40% further than the row above. Afterwards the list price applies again on its own.
Estimates, not an allowance in questions. Two things move them most. Thinking: the assistant thinks before answering by default, at the High level, and thinking is charged like any other output — turning it down or off under ⚙ Settings → Assistant stretches the budget several times further. Portfolio size: a question about fifty holdings costs more than one about five. Any model stays selectable on the free budget; the pricier ones simply drain it faster.
Two daily caps protect the shared budget on the hosted app: a total for all free use, and a limit on how many new free budgets one network can start. When either is reached the chat says so; your own provider is not affected.
One more limit: how fast you can ask
Separate from the budget: on the hosted app the assistant answers up to 28 requests a minute per network address without an account, and 40 per account once you are signed in, so sharing an office connection with other users cannot slow a signed-in session down. That counts every request made on your behalf — your questions, the proactive tips, the ✨ insight pills and the one-click reports. Normal use does not reach it. If you do, the chat says so in a line of its own, nothing is spent, and the same question works again a few seconds later; the message is worded differently from the one for a spent budget so you can tell them apart.
Locking your API key down
A key you paste into Settings is used only for your own portfolio's calls. You can also restrict the key itself at your provider, so it is useless anywhere else:
- Spend caps and usage limits. The OpenAI and Anthropic consoles, and most aggregators, let you cap a key's monthly spend or rate.
- Scoped keys. Where the console offers project- or scope-restricted keys, create a dedicated key for Fortunest.
- IP allow-listing. Where the provider can restrict a key to caller addresses, allow the app server's address. The hosted app shows it directly under the API key field in ⚙ Settings → Assistant, with a Copy button. This applies to the hosted app only: the desktop app calls the provider from your own computer, so prefer spend caps there.
The two model slots
- The conversation model answers your chat questions and writes the one-click reports. Thinking and effort controls appear where the model supports them (thinking on, high effort by default) and apply to every message.
- The Fast model (same model list; thinking is always off) runs everything else — background insights & search: proactive comments, the ✨ AI insights, ✨ smart search in the finder and Explore, the semantic holdings filter, and article and filing summaries. Metric and top-N holdings filters (
"top 5 holdings","my best PEG holdings") are answered locally without calling any model.
Both picks are remembered per provider. Changing either model or its thinking settings clears the AI caches, so one model's output is never served as another's.
Setting up a local model from scratch
With Fortunest desktop and the model running on the same computer, local inference has no per-token charge and the model request stays on the device. Hosted access through a tunnel is different: it does not make the hosted portfolio local. The separate online services are listed in desktop privacy. You need three things: a server app (LM Studio or Ollama, both free), a model your hardware can hold, and a large enough context window.
Context window: at least ~32 000 tokens, comfortably 50 000+. Every question travels with the assistant's instructions, your portfolio context and the conversation so far; the context figure in the chat header shows your own number. Many tools load models with a 4 096 or 8 192 default. That is far too small: the figure turns red and the assistant refuses the message with a line that says what to change. This is the most common reason a local setup connects but fails.
LM Studio, step by step
- Download LM Studio from lmstudio.ai (macOS, Windows, Linux) and open it.
- Get a model from the search tab. Good starting points:
openai/gpt-oss-20bon a 16 GB+ machine, or an 8B-class model (Llama, Qwen) on 8–16 GB. - Load it with a bigger window. In the model's load settings raise Context Length to at least ~32 000, ideally 50 000+. It is a per-load setting: reload the model to apply a new length. The app picks up the new value within a minute.
- Start the server: Developer → Start server. The address beside Reachable at —
http://127.0.0.1:1234by default — is where Fortunest looks, so on the same machine there is nothing to enter. - In Fortunest: ⚙ Settings → Assistant → LM Studio. Pick the conversation and fast models from the dropdowns and press Save.
Ollama, step by step
- Install from ollama.com (macOS, Windows, Linux;
brew install ollamaalso works). It runs in the background athttp://localhost:11434, the default the app probes. - Get a model either from Fortunest — Settings → Assistant → Ollama offers a curated list with disk sizes — or on the command line:
ollama pull llama3.1:8b. - Context is handled for you: the app requests a 32 768-token window on every Ollama call. Your machine still needs the memory to serve it.
- Pick the models in ⚙ Settings → Assistant and press Save.
A hardware reality check
The model file has to fit in memory (RAM, or GPU memory where you have it), and a big context window adds to that. For quantized models: an 8B model wants ~6–8 GB free, a 20B-class model ~13–16 GB, 30B+ upwards of 20 GB. If loading with a 50k context fails or answers crawl, try a smaller model before a smaller window.
The Server field in Settings → Assistant is for the exceptions: a non-default port, a model on another machine on your network, or the hosted app reaching your machine through a tunnel.
Local models with the cloud app — advanced
Advanced users only — at your own risk. This exposes a server running on your computer to the public internet for as long as a tunnel is up. Read the risks first. The safer way to use local models is to run Fortunest itself on your machine with the desktop app: the app and the model share localhost and no tunnel exists.
The hosted app calls your model from our server, so it needs a public address that forwards to your computer: a tunnel. When you pick Ollama or LM Studio on the hosted app, a dialog asks you to confirm; the Server field then takes the tunnel URL. Only publicly routable URLs are accepted.
Option A — ngrok
- Create a free account at ngrok.com and install the agent:
brew install ngrokon macOS,winget install ngrok.ngrokon Windows, or the download on their site. - Connect the agent to your account once:
ngrok config add-authtoken YOUR_TOKEN. - In LM Studio, load a model and start the server (Developer → Start server, port 1234).
- Start the tunnel:
ngrok http 1234. It prints ahttps://…ngrok-free.appaddress. - In the app: ⚙ Settings → Assistant, pick LM Studio, confirm the warning, paste the address into Server and press Save. Your loaded model appears in the dropdowns.
- When you are done, stop the tunnel (Ctrl-C). The next start prints a new address; paste it over the old one.
Option B — cloudflared (no account needed)
- Install the agent:
brew install cloudflaredon macOS,winget install Cloudflare.cloudflaredon Windows, or the package from Cloudflare's downloads page. - Start LM Studio's server as above, then:
cloudflared tunnel --url http://localhost:1234. It prints ahttps://…trycloudflare.comaddress. - Paste that address into Settings → Assistant → Server (after picking LM Studio and confirming the warning) and press Save.
Using Ollama instead of LM Studio
The same, on port 11434 — but Ollama refuses requests whose Host header is not local, so tell the tunnel to rewrite it:
ngrok http 11434 --host-header=localhost:11434, orcloudflared tunnel --url http://localhost:11434 --http-host-header localhost:11434
Skip the OLLAMA_HOST=0.0.0.0 OLLAMA_ORIGINS=* advice you may find elsewhere. 0.0.0.0 binds Ollama — which has no authentication — to every network interface, so anyone on your Wi-Fi can use it; ORIGINS=* lets any website you visit call it from your browser. The host-header rewrite needs neither.
The risks, and how to shrink them
LM Studio's and Ollama's servers have no authentication: while a tunnel is up, the URL is the credential.
| Risk | What it means — and how to shrink it |
|---|---|
| Compute theft | Anyone who learns the URL gets free inference on your hardware, and LM Studio's management endpoints let them load or eject models. Keep tunnels short-lived: both vendors issue a fresh random URL per start. |
| URL leaks | The addresses are not practically guessable; the realistic leaks are screenshots, screen shares and pasted logs. Treat the URL like a password. It is also stored in your session's settings, where the deployment's operator could see it. |
| Traffic visibility | TLS ends at the tunnel vendor's edge, so ngrok or Cloudflare can technically see the traffic, including the portfolio context sent to your model. |
| Server bugs | These are development servers, not hardened internet services — Ollama has had remotely exploitable bugs (CVE-2024-37032). Keep them updated and the exposure window short. |
| Unexpected traffic | LM Studio's server console prints every request. Traffic you did not cause means someone has the URL: stop the tunnel, which ends their access. |
A URL-holder cannot get your portfolio data, your chat history, your app session or files on your disk (short of an unpatched server vulnerability). Basic-auth on the tunnel (ngrok http 1234 --basic-auth …) does not work: the app sends no credentials and refuses URLs with credentials embedded in them.
Chatting
The question box reads Ask anything…: your portfolio, any other asset ("How has NVDA done this year?"), your strategy, how the app works, a chart or table drawn from your data, a pasted screenshot or an attached PDF, or a dashboard command ("Open the Risk card"). Type / into the empty box for the verb list: every dashboard action the assistant knows, with its parameters and one line of what it does, narrowing as you type. Enter or a click fills the box with verb {parameters} for you to complete in words. A slash later in a sentence ("sell 3/4 of it") is just a slash.
Before a model is connected the pane prints the reason once — the same sentence as the status dot beside the title, e.g. "No assistant configured — add an API key or a local model under ⚙ Settings → Assistant" — with a Set up the assistant button, and the Learn card below it. Once a model is set up, the conversation opens with the ✨ report cards and example questions that follow the tab you are on: "Show my top 5 holdings" on Portfolio, "Show the biggest movers this week" on Explore, "Which lots become tax-free within 12 months?" on Taxes. One more, lit differently, is built from your largest holding ("What is driving NVDA this week?"). If a configured model cannot answer right now (switched off, free budget used up, server down), the reason sits above the cards and chips, which stay dimmed.
Walkthroughs. The Learn card holds two chips, Dashboard tour (Beginner / Intermediate / Advanced) and 🌱 Investing Foundations. The course is 25 steps about investing itself, for people who have not made their first trade. It needs no data and loads none, and skips what you say you have already covered. Its charts use 30 years of market history bundled in the app, and the Simulation tab does the arithmetic on your own numbers. It names common approaches without recommending any: the app takes no house view. It ends by writing your plan into ⚙ Settings → Strategy and your Investing notes, then offers the Beginner tour. Every walkthrough is pregenerated and plays without a model; only questions asked along the way need one. A chapter that needs Taxes (Pro) or Social (a signed-in account) without that access runs as Guidance only. Ask "restart the investing course" and the assistant starts it again at step 1 (tutorial {op: start, level: foundations}). The Learn card's chips, ticks and retakes are described in Take the tour.
👀 What you're looking at. Every message carries your live view (tab, currency, filters, any open drawer or dialog, the simulation you ran), your last 60 actions, the cards on screen and what you are looking at, so you can ask about "this position" without naming it. A one-line chip under the conversation shows the last part: the tab, the drawer over it, the card under your pointer and the asset, term, chart point or value inside it, plus any filters on that card (“Holdings (filter: “nvda”; type: Crypto)”). Each question also carries what you rested on in the last 3 minutes, so "What do you think about this company?" needs no naming; on a chart, the hovered day itself is passed on. Hover the chip for every part in full. The chip shows only while a model is connected; on a phone it holds the last thing you tapped. Other people's words are never passed on: on the Social tab the assistant learns only what kind of thing you are looking at (“a post in the feed”). An agent connected over MCP is not told where your pointer is. To see everything that is sent, open the Context window.
- Ticker chips and the asset card. Symbols in replies and in your own messages become chips; so do company names written the way you write them (nvidia, Amazon, Google). Click a chip to open the asset's details from any tab. Hover it for the asset card: the company's logo, ticker and full name; a one-week mini chart; the price in the asset's own trading currency; today's move (or the last session's) and the week's move, each with its percentage. For a holding it adds the position's value in your display currency, its share of the portfolio and its total P/L with return. A line the data cannot support is left out. The same card opens over Value per asset, a friend's category pie (share and published return only) and the assets on the Trades tab; on touch screens a tap opens the details instead. Ordinary words stay text: a short ticker that is also a word (A, ON) chips only where marked as one (“Agilent (A)”), and an asset you neither hold nor watch chips only in capitals or as a name mid-sentence.
- Drag a chip onto your Favourites: onto a Favourites list tab (into that list, starred on the way), onto the + (a new list that starts with it) or onto the Favourites table. It works from every tab, with the chat docked or floating; a note confirms the drop (NVDA added to “Tech”.). Clicking the chip still opens the asset. Chips in reports, Social posts and Super Investor filings drag the same way. On a touch screen, tap the chip and use the ★ in the asset's details.
- Web addresses show as plain text you can select and copy, never split by a ticker chip and never clickable, since the assistant may be repeating one from a news headline. Links into this manual stay clickable.
- Charts: interactive Vega-Lite charts render inline, expand to a full-window view, zoom by dragging a rectangle and download as PNG or JSON. Tooltips show text only, and a chart never loads data from the web.
- Formatting: tables (a wide one gets a ⤢ to maximize it), headers, lists and code blocks, with ticker chips working inside them. Formulas are typeset:
\[ … \]or$$ … $$on their own line,\( … \)inside a sentence. A single dollar sign is always money. - It knows this manual: ask how something works ("how do I export my Revolut history?") and it looks up or searches these docs, links to them (hover a link for a preview) and can show the manual's screenshots below the answer.
- Attachments: 📎, paste, or drag files onto the question box. Images need a multimodal model (a picture-frame badge next to the model name); PDF and text files work with any model, as extracted text. A file may be up to 20 MB and a PDF up to 200 pages; the model reads the first 60 000 characters of its text. A scanned PDF with no text layer says so rather than arriving empty, and a file that cannot be read is skipped with a note while the others attach. A paste from a spreadsheet keeps the cells as text. A link dragged onto the box is typed into your question. Click a sent image to open it full-window, with zoom and pan.
- Keep the conversation: the ⇩ button beside ⟲ saves the chat as a self-contained HTML file (
fortunest-chat-2026-08-13.html) or opens the print dialog for Save as PDF. Both are made in your browser and need no internet connection to open. They carry only what the bubbles show, with charts as pictures and asset chips keeping a hover card captured when you saved. No hidden portfolio context, model or provider name, token counts, account details or keys are included. - Copy and timestamps: Copy at a reply's top-right corner (on hover; always shown on a phone) copies its raw markdown; a code block has its own Copy. Click a reply to see when it was generated and what it cost (“· 4 210 in / 380 out” tokens, when the provider reports it); click your own question for its Sent time and its 👀 chip.
- Rewrite a sent question: hover it and click ✎ to edit it in place. Sending the changed wording restarts the conversation from there, dropping that answer and everything after it; Escape or Cancel leaves the history as it was.
- Drafts are kept. A half-written question survives closing or minimizing the chat, switching between floating and docked, changing tab and reloading the page in the same tab. Only sending clears it. Staged files survive closing the chat but not a reload.
- Enter sends, Shift+Enter breaks a line. ⟲ resets the conversation.
Ask AI — from a tooltip or an alert
Every ⓘ tooltip that explains something (and other explanations that open on hover, such as the Fair value card’s model names), and every row of Insights & alerts, carries an Ask AI button; hover it and it says “Ask the AI assistant more about this.” Press it and the assistant opens, the question lands in the current conversation as your own message, and the answer starts at once. The question quotes what you were reading (Explain more about this: “Cash — …”), with the time for an insight or alert. Social and consensus-shift alerts carry other people's words, so they are asked about by title and time alone. Asking does not tick off the insight or mark the alert read. Press Ask AI again while an answer is still coming and the question waits its turn: up to three wait and are sent one after another; a fourth goes into the ask box for you to send later. If the assistant cannot answer right now, the chat opens on its explanation and nothing is sent.
The tooltip stays open while your pointer rests on it. By keyboard, Tab to an ⓘ, Tab again to Ask AI, Enter to ask. On a phone, tap the ⓘ, then the button. Tips that only read out a value (a chart's day, a heatmap cell) have no Ask AI.
Asking “what should I buy?”
The answer depends on what the assistant knows about you. With a portfolio and a written strategy (⚙ Settings → Strategy) it recommends from your data, honouring your thesis, horizon and risk tolerance. With a portfolio but no strategy, it reads your trading pattern back to you (one fund bought at a steady rhythm looks like passive dollar-cost averaging; many picks and rebalances look like active stock-picking), confirms with a couple of questions, and saves the result as your strategy only once you agree.
On an empty portfolio, or once you have started the Investing Foundations course, it recommends no tickers at all. It sketches the decision framework (owning vs lending, whole-market index funds vs picking stocks), then asks up to four profiling questions, one per reply, with clickable answer chips (or type your own answer). Only what you answer is saved: your horizon, your risk tolerance, a one-sentence strategy in your own words and a monthly amount for the trade plan. It then points you at the Explore tab and the simulator with your own numbers; any fund it names is an illustrative example of a category, never advice to buy it.
The dock
On a computer the assistant opens as a full-height side pane on the right that pushes the dashboard left; drawers, dialogs and maximized charts stop at its edge, and the dashboard scrolls in its own column, with its scrollbar between the dashboard and the chat. A window control turns the pane into a floating window in the bottom-right corner (drag its top-left corner to resize). Minimize folds it into the launcher disc in the bottom-right corner; its sparks pulse faster when a reply is unread, and a small light shows whether the assistant is connected. Your floating or minimized choice lasts for the browser tab; a new tab opens docked again. The exception is while the Get started card is showing: a new tab then starts minimized until you have used the assistant (sent a question, started a walkthrough or pressed the card's Open the AI assistant chat).
Narrowing the window narrows the pane down to its minimum, then the dashboard. In a window too narrow for both (under 862 px) the pane folds into its disc and re-opens as the floating window; widen the window and the pane comes back, unless you minimized it yourself. On a phone the assistant is a bottom sheet opened from the disc in the bottom control panel; the panel-height button beside × drops it to a short height so you can scroll the page behind. When the assistant opens an asset's details, or you tap a ticker, the sheet drops to its short height until the details close.
What the assistant sees
⚙ Settings → Assistant → Assistant context is a checkbox per part of what a question carries: checked rides along with every message; unchecked is left for the assistant to read on demand with a focused read, costing nothing until a question needs it. One setting for the whole app, the same on every plan, applied when you press Save.
- Checked to begin with — Investing strategy, Investing notes, Memories, Macro & rates, Recent trades.
- Investing notes — locked on, marked (always): your notes ride every message.
- Unchecked to begin with — Upcoming events, Favourites, Recent news, Trades & recommendations, Product docs index. Tick one you ask about constantly to save the read. Leaving Product docs index unchecked keeps the instructions short — the assistant searches the manual instead — which matters on a small local model.
- Full portfolio snapshot — the complete snapshot instead of the core digest. It includes every section above, so while it is checked the other boxes have no effect; by far the most tokens per question.
- Trade window — on the Recent trades row: months of your own trades that travel (0–24, default 3; 0 means none).
The line under the list says whether the model in play can call tools. If it can, anything unchecked is a question away. If it cannot, everything is sent with every message regardless of the boxes. Such a model asks for a read by naming it in its reply. Ollama and LM Studio report whether a local model can call tools; only when the server says nothing does the app judge by the model’s name.
The Context window behind the chat header’s context: figure lists each section with its token estimate and splits the last message’s input into instructions, portfolio core, tool results and conversation history. On demand means not sent (its zero is not “empty”) and shows what a read would cost; · carried in the digest means the tokens are counted on the digest’s row.
Focused reads
A focused read is a small, named, read-only request for exactly the slice a question needs, taken from the same valuation as the rest of the answer, so nothing can disagree. The assistant reaches for the smallest read first and treats get_context (everything at once) as a last resort. Agents connected over MCP get the same reads. Every result names its currency, and a long list comes back a page at a time. Company reads take several tickers in one call — up to 25 (headlines 20, calendar and insider filings 30) — and name any ticker that has no data.
| Read | What it returns | What it can be asked for |
|---|---|---|
get_core (core digest) | The whole portfolio in about 4,000 tokens: totals with the per-account split, returns, the six Risk-card headlines, the ten largest positions, allocation against targets, plan totals, the next five events, warning counts, strategy, rates, recent trades, plan limits, memories and notes. When it is too big, the longest lists are shortened and named. | include (which optional blocks to keep: memories, notes, strategy, macro, recentTrades) |
holdings, holding | Positions, ten at a time with the total count; one position in depth as its drawer shows it — fundamentals, lots, news, insider filings, events, correlations, warnings, price history, its description and holding checks. A ticker you neither hold nor watch is answered “not held or watched”. | sort (weight, value, day change, gain, loss, name), order, filter (account, asset class, category, sector, region, currency, symbols, minimum weight), fields; symbol/symbols, include, range |
watchlist, news | Favourites with price, day and month move, rating, fair-value gap and whether you own them (prices in US dollars); headlines for your holdings, favourites or named symbols. | list, sort, fields (core, full); scope, symbols, since (days, default 7), limit per symbol |
allocation, risk | One split, largest first, beside your targets (an unmet target is raised, holdings awaiting data flagged); the risk figures in pieces, including the diversification score and the Fingerprint. | dimension (asset class, sector, region, currency, size, type, account), depth, withTargets; include (metrics, drawdown, correlations, diversification, benchmarks, FX attribution, fingerprint), range (all, 1y) |
chart_values | The value chart’s lines on a day (today by default), from the series the chart draws: your total, your other lines and each benchmark, with change, return and the highest benchmark named. In the chat it follows what your chart shows. | date (up to six), benchmarks, lines, from |
trade_candidates, get_trade_plan, recent_trades, money_flow, tax_lots, tax_free_summary, sell_plan | The Trades and Taxes cards: trade candidates with their company rating, leading criteria and the plan’s reason; the Trade plan; Recent trades row for row; Money flow per period; maturities with days to tax-free, lot lines, gains and tax per year; what is tax-free now; which lots to sell to raise an amount. | side, driver, owned, symbols, fields (extra fundamentals on up to ten candidates — “forward P/E against PEG for the recommended buys” is one read), why; plan include; trades since, sort, query, type, from/to, detail; period (week, month, year), count; view (maturities, openLots, annual, ytd), year; amount |
events, insider_activity | The forward calendar — earnings, dividends, central-bank dates, lots turning tax-free; earnings alone answer one row per company with next and last report, beat or miss. Open-market insider trades from cached SEC Form 4 filings for what you hold, net in US dollars. Nothing is fetched; filings are context, never a signal. | kinds (earnings, dividends, macro, maturities), days, symbols, scope (holdings, watchlist, both), past; since |
simulation | What your simulations start from (value, historical return and volatility, horizon, risk, budget). The assistant never runs a projection; a historical replay is kept for an hour and comes back with its result and warnings. | include (summary, whatif, dcaGoal, backtest), backtestId |
warnings | Data-quality notices — unread rows, assumed balances, missing prices or rates — with counts per kind. | kind, symbol |
market_lookup, price_on, search_assets, read_market_list | Any listed asset: quote, news, earnings; prices or exchange rates on past dates; asset search; an Explore list. | parts; date, currency, fx; query; list |
docs_search, docs_lookup, list_notes, list_memories, get_context | This manual; your notes and memories; the full context by section. | query; labels; sections |
Limits. Reads run on our server inside the same request, all reads of one reply at the same time. A question that needs a second look gets one, up to three rounds, then the assistant answers with what it has; one reply makes at most eight reads per round and twenty-four in all. Each result is capped (the core digest gets twice the room) and a cut list says how to ask for the rest. Reads only read: anything that would spend or change something is refused. Dashboard changes are actions.
Head start. With a model that can call tools, the quick pass that reads every message first may name up to four reads a data question needs, so their results arrive with your question and “What are my top 5 holdings” takes one round instead of two. A wrong guess costs nothing. A question the app asks for you can bring its read along itself: the demo intro card’s Ask the AI about this portfolio sends “Which of my holdings made the most money?” with your holdings sorted by money made, so that answer skips the quick pass and starts on the read straight away.
In the conversation. Each read appears as a row above the answer and fills in with what it found — “Read your holdings (top 5 by weight)” — before the reply starts. Consecutive reads share one bubble with the ⚡ icon and a count; click its header to fold it, or on eight or more rows use the round Collapse actions button at the bottom right. A failed read keeps its error visible. The rows stay with the answer and in a saved conversation, and are never sent back to the model.
Driving the dashboard from chat
Type a command such as "Go to allocation" and the assistant opens that card and comments in the same reply with the numbers it shows. The starter questions under the report cards change with the tab.
| You type | What happens |
|---|---|
Navigate & reveal
|
Opens the matching card: Allocation, Risk, Dividends & income, the Events card’s Upcoming section or the Trading activity card on Trades, and comments on it. |
Filter holdings
|
Sets the holdings table’s smart filter. Top-N and metric queries are answered by rules and come back ordered; descriptive ones (“tech stocks”) run on the fast model. |
Control the chart
|
Sets the value chart’s view, range and benchmarks, then summarises how the portfolio did over that window. |
Run simulations
|
Runs the what-if trade or the goal seek on the Simulation tab and summarises the result and its assumptions. |
Explore the market
|
Opens the Markets table and its eight lists, or a heatmap over the list and window you name: the Market heatmap, also over your watchlist or one named list, or the Portfolio heatmap. |
Look up any company
|
Fetches live figures for any company, held or not: quote, ratios, score, headlines, earnings and the results filing downloaded from the SEC. Gaps are named with their reason; European listings, crypto and most ETFs have no SEC filing. |
Quick actions
|
Opens the asset’s detail drawer with a summary, stars a stock into your favourites, or applies the theme or display currency. |
Change settings
|
Changes ⚙ Settings values by name, several at once and all-or-nothing, as Save would. Out-of-range values are refused; while ⚙ Settings has unsaved edits the assistant declines. |
Ask where something is
The assistant knows every tab, card, chart view and ⚙ Settings section. Ask "where can I see my drawdown for this year?" and it answers with a link to the place that fits: Portfolio value over time › Drawdown for the whole portfolio, Asset details › Drawdown for one asset. Say "take me there" and it goes itself, including to a ⚙ Settings section ("where do I set my allocation targets?"), which it never changes.
Custom assets by conversation
Assets added by chat appear under ⚙ Settings → Data.
- Manual assets:
"add my apartment in Prague, bought 2019 for 200k EUR". Missing name, date or price is asked for. The price is the value on that date, grown at the optional yearly change; for today’s worth give today’s date. - Market-priced assets:
"add 10 shares of RHM.DE", priced live from then on. Money works too:"my bitcoin is worth 1,000,000 CZK"(today) or"I put 50,000 CZK into SAP in October 2021"(paid then), converted to units at that day’s close and exchange rate. Date defaults to today, price to that day’s close. - Statements: positions read from an attached file or a screenshot are added one asset per line. To import a statement as one reviewed account, use read with AI.
The app picks the listing when the choice is settled: one you hold or watch, then a coin’s -USD pair (“BTC” is BTC-USD, not the ETF of that ticker), then the exact ticker. The assistant asks only when the matches are genuinely different assets, and its reply lists every default it applied so you can correct it. Each asset joins a named group that fits, or a new one.
"change the buy date for SOL and AAVE to 2021-09-01" edits hand-added positions, only the fields you name; a price you never typed follows the date. "exclude my Prague flat" or "turn my Kraken account back on" switches an asset or data source without deleting it. Broker imports stay managed by their files. Several changes in one reply share a single portfolio recalculation.
"set the category of all the cryptos I own, except Bitcoin, to Altcoins" sets the Category column of any holdings, imported ones too, all or nothing: an unknown or ambiguous name, or a category tied to another sector, changes nothing. Other or empty clears it.
What you can rely on
- No approval step: the assistant’s own actions run at once; only connected agents wait for your approval.
- No silent failures: an action that cannot run is named with its reason, and the question is answered in text.
- Real numbers: commentary uses the data on the card.
- Actions become context: they join your recent actions, so follow-ups know what is on screen.
- Visible results: each chip shows ✓ or ✕ with the reason as tooltip; the card it drove flashes.
- Undo: edits, include switches and category changes carry an inline Undo that restores exactly the fields changed; remove an added asset under ⚙ Settings → Data. A connected agent’s Undo refuses to overwrite a newer choice, such as a model picked since.
A turn that goes wrong
- Retry on a failed turn’s error line resends the question.
- Stop replaces Send while an answer is written; the partial answer stays, marked — interrupted.
- A refusal reads “The model declined to answer that.”
- A cut-off answer ends “— cut off (length limit)” with Continue.
- A long conversation is shortened to the model’s context window (after the slash in the header’s
context:figure), oldest turns first; the figure turns amber near the limit, red past it. Windows: Claude 1M tokens (Haiku 4.5 200K), OpenAI 272K, Ollama up to 32 768, LM Studio the length the model was loaded with, OpenRouter its catalogue’s figure. - Too small a window: nothing is sent, and a red line says what to change: a model with a larger window under ⚙ Settings → Assistant → Assistant model, a longer LM Studio context (Ollama runs with up to 32 768 tokens), or less Assistant context.
- A provider failure (overload, rate limit, bad key) shows an error with Retry and a
ref, e.g. “[ref 3f9a1c2b]”; quote it when you report a problem. “The model returned an empty answer.” means exactly that.
The complete action reference
Everything above runs on 54 typed chat commands; a request that fits none gets a plain-text answer. A skill can name one (“use the show_card action with card allocation”). ? marks an optional parameter. Fetched results go back to the assistant, which answers from them.
Actions: navigate & reveal
| Action | Parameters | What it does |
|---|---|---|
show_card | card: Portfolio allocation, risk, events, income, holdings, chart, insights, value, cash, pl, fundamentals, drift, assets; Trades recommendations, trade_plan, recent_trades, trade_flow, activity; Taxes trades_summary, tax_lots, sell_optimizer, maturing_lots, annual_gains, transfers, manual_lots. expand? (false collapses), collapse? (cards), section? (upcoming, events), sector?/category? (assets; null resets) | Opens the card on its tab; dial tiles and drift open as a dialog, insights opens the alerts bell. sector/category open the Value per asset pie on that slice. |
show_trades | — | Switches to Trades. |
show_taxes | year? | Switches to Taxes; a year opens its tax breakdown. |
show_social | pane?: feed, friends; handle? | Switches to Social; a handle opens that friend’s portfolio. |
show_simulation | card?: whatif, simulator | Scrolls the Simulation tab to that card. |
open_settings | tab: general, data, strategy, trades, taxes, social, alerts, assistant, about; section? (a label, e.g. "Allocation targets") | Opens ⚙ Settings there; an unknown section is refused with the real ones. |
explore | table: top, movers, favourites, heatmap, macro; list?: top, movers, sp500, nasdaq100, dow, ftse100, dax, nikkei225 or a watchlist name; heatmap: mode? (market, favourites, portfolio), range? (1D…10Y, Pos), metric? (change, drawdown) | Shows a Markets list, your watchlist, the Macro card or a heatmap: mode picks the grid (omitted: the one on screen), range its window (Pos = since first held, portfolio only), metric what tints it. Weekly movers are a heatmap at 1W. |
sort_table | table: holdings, top, movers, favourites; column (e.g. valueEur, dayChangeEur, plEur, weight, lastTraded; Explore dayPct, marketCap, forwardPE); desc? | Sorts like clicking the header. |
expand_trades | target: lots_row, lots_all, recs_why, recs_all_buys, recs_all_sells, weights; symbol (for lots_row, recs_why); on? | Expands or collapses that element. |
open_asset | symbol?; section?: drawdown | Opens the detail panel of any listed symbol, held or not; unknown tickers come back with near misses. |
open_breakdown | which: value, pl | Opens the Portfolio value breakdown or Total P/L views. |
Actions: chart, holdings filter & the tutorial
| Action | Parameters | What it does |
|---|---|---|
set_chart | Any of mode? (overall, companies, returns, drawdown); range? (1D, 1W, 1M, 3M, 6M, YTD, 1Y, 5Y, All); from? (a date, a year, "all" clears); benchmarks_on?/benchmarks_off? (sp500, world, nasdaq, bonds, gold, usd, btc, a custom benchmark; off "all"); lines_on?/lines_off? (total, an account, diversification, inflation, type:Crypto, sector:Technology, category:Semiconductors) | Adjusts the value chart. range and a from date exclude each other. |
set_as_of_date | date: "YYYY-MM-DD" or null (now) | Time travel: the Portfolio tab at that day’s close, with its figures. Days outside the price data are refused with the earliest valid one. |
filter_holdings | query; "" clears | Sets the smart filter (“top 5 holdings”). It reads a description; thresholds are filter_holdings_columns. |
filter_holdings_columns | filters: list of {column, min?, max?, only?, exclude?}; clear?; period? (1W, 1M, 3M, 6M, YTD, 1Y) | Sets several column funnels, all of which must hold. column: a Holdings column key or heading, or dayChangePct, periodChangePct, currentDrawdown, ter. min/max are inclusive in the column’s units (2 = 2%); category, sources, type, sector take only/exclude. clear first removes every filter. E.g. {clear: true, period: "1M", filters: [{column: "periodChangePct", min: 2}, {column: "plEur", min: 0}]}. |
tutorial | op: start, next, stop; level?: beginner, intermediate, advanced, foundations | Drives a walkthrough in the chat, including the Investing Foundations course; start restarts a finished one. |
Actions: search, live market data & this manual
These fetch data and leave your view as it is; your own figures come from the focused reads.
| Action | Parameters | What it does |
|---|---|---|
search_assets | query; smart? | Searches the topbar finder; hits (symbol, name, exchange and type) return to the assistant, which picks the ticker itself unless the matches are genuinely ambiguous. |
market_lookup | symbol or symbols (up to 25); parts?: quote (default), news, earnings | Live data for any company: ratios, headlines, earnings dates, the SEC filing. |
price_on | date (up to 8, or "today"); symbol? or symbols?; currency?; fx? (up to 4, e.g. USD/CZK) | The close on a given date, at that day’s exchange rate; a closed market answers with the previous trading day and says so. |
docs_lookup | labels: 1–5 "page#anchor" | Reads sections of this manual. |
docs_search | query; limit? 1–8 | Searches the manual and website. |
Lookups in one reply share one size limit; news and filing text give way first, and the assistant says what was cut.
Actions: favourites & watchlist
| Action | Parameters | What it does |
|---|---|---|
toggle_favourite | symbol or symbols; on; list? | Stars or unstars; with list, adds to or removes from that named list, created if new. |
favourite_list | op: create, rename, delete; name; newName? | Manages named lists; deleting one unstars nothing. |
Actions: simulation
None of these places an order.
| Action | Parameters | What it does |
|---|---|---|
whatif | symbol; side: buy, sell; one of quantity, amount (display currency; amountEur deprecated), fraction (0–1); or trades, a list of those | Runs the what-if card; several trades combine, sells first. |
dca_goal | goalAmount; targetYear, targetMonth ("YYYY-MM") or targetDate; assets? (symbols) | Runs the goal seek. |
set_simulation | Any of horizonMonths? (1–360), lookbackMonths? (6–240), mode? (dca, goal, none), assets? ({symbol, amount?, days?} rows), sells? ({amount, days?} rows, or false), benchmarks?, diversification?, uncertainty?, lines? (withDca, withoutDca) | Adjusts the Future simulator, its P10–P90 band and lines. On an empty portfolio DCA and goal seek run from zero when you name the assets; Projection and the backtest show how to add a portfolio instead, and withdrawals are refused. |
backtest_trades | interval? ({unit: months, days; value}, monthly default); weights? (0–100 for quality, valuation, fairValue, analystUpside, momentum, drawdown, dividend, group_consensus); transferPolicy?: strict (default), cashEquivalent | Replays last year’s recommendations in Historical mode from dated prices and filings, with today’s settings. The answer gives actual and passive results, coverage and exclusions; cashEquivalent, used only when you choose it, leaves acquisition dates and tax costs unknown. |
Actions: trades & taxes
Tax answers are orientation, not advice. Explaining a plan, the assistant sees the allocator’s limiting reason: a small holding’s limit, which a larger total sell budget does not lift, versus a budget too small for the order count.
| Action | Parameters | What it does |
|---|---|---|
set_asset_target | name (symbol); pct 0–100, 0 removes | Sets an allocation target. |
set_sector_target | name (sector); pct | The same, for a sector. |
tax_free_summary | symbol? | Tax-free sellable value, next maturity, proceeds against the small-sale limit. |
sell_plan | amount; symbol? | Runs the Sell optimizer. |
add_manual_lot | symbol; date; qty; unitCostEur (EUR); source?; note? | Adds an opening-balance lot. |
transfer_action | symbol; op: confirm, dismiss, undo | Settles a suspected transfer. |
Actions: adding & editing data, notes
| Action | Parameters | What it does |
|---|---|---|
add_manual_asset | name; date; price; currency?; apy? (% a year); assetClass? (Cash = cash); category?; group? | Adds a manual asset; price is its value on date, compounded at apy. |
add_market_asset | symbol or query; quantity, value (today) or cost (paid), or quantity with price/cost; currency?; date?; price?; category?; group? | Adds a market-priced position. Units with a value must match today’s close within 10 %, a price must equal cost ÷ units within 2 %. A currency pair (EURCZK=X) adds cash. |
edit_asset | symbol, name or id; quantity? (new total), date?, price?, category? | Edits a hand-added asset; for a manual one price is the value on its date. |
toggle_asset | symbol, name or id; on | Its include tick; a bare symbol switches every lot. |
toggle_account | name; on | The same for a whole data source. |
set_category | symbols (up to 200); category (48 characters) | Sets the investing category of any holdings. |
sync_connector | name? (account or platform) | Syncs connected accounts now. |
add_note | text (500 characters) | Saves an investing note 📝; 20 at most. |
save_memory | text (300 characters) | Saves an assistant memory 🧠; 50 kept. |
Actions: social
These need a signed-in account with Social on. Friends’ figures are relative, never money.
| Action | Parameters | What it does |
|---|---|---|
friend_portfolio | handle; brief? | Reads what one friend shares; up to four per reply. |
group_portfolio | group: friends, investors; weighting?: equal, aum; offset?, limit? | Compares a group portfolio with yours (“what do my followed funds hold?”); gap = your weight minus the group’s. Investor groups need Pro. |
super_investor_portfolio | name | A followed Super Investor’s latest 13F: quarterly, up to ~45 days old, US long positions only. |
follow_super_investor | name; off? | Follows or unfollows; Free follows two. |
social_feed | limit? 1–10 | Reads the friends’ feed. |
social_requests | — | Lists friend requests; answer them on the Social tab. |
publish_snapshot | — | Publishes now. Publishing is otherwise automatic. |
Actions: settings & appearance
| Action | Parameters | What it does |
|---|---|---|
set_theme | mode: dark, light, auto | Switches the theme. |
simple_view | on?; cards?, the full ordered list from value, cash, pl, fundamentals, chart, heatmap, assets, allocation, risk, income, events, holdings | Sets the Simple view. |
set_currency | code (30 currencies) | Switches the display currency and converts the trade plan’s amounts, naming both; refused when no rate is available. Tax thresholds keep their own currency. |
set_model | model; fast? | Switches the chat or fast model. |
set_setting | key, value and, per key, source, scope, days, model, cond, metric, kinds; or settings, a list | Saves as ⚙ Settings does, a batch all-or-nothing. Keys below. |
| Group | set_setting keys and values |
|---|---|
| General | quote_refresh_minutes (15, 30, 60, 120, 240, 480, 720, 1440); favicon_dot; proactivity (off, minimal, active); context_trade_months (0–24); context_sections (true/false per section: strategy, notes, memories, macro, recentTrades, events, favourites, news, tradesDigest, docs, portfolio) |
| Strategy | strategy_text; strategy_horizon_years (1–50); strategy_risk (low, medium, high; swaps in that level’s preset unless you wrote your own thesis) |
| Trade plan | buy_amount, buy_count (1–10), buy_min_order, sell_amount (0 = none), sell_count (1–10), sell_min_order; buy_candidates_from (owned+favourites, owned, favourites — the Buy candidates from setting; the sell side is always what you hold) and buy_candidates_list (a list name or all) |
| Planner | max_company_weight_pct (0.01–100, null clears); allocation_mode (ratings, targets); planning_targets ({funds, crypto, stocks, cash}, totalling 100%). Invest is new money; Sell is independent. Rules. |
| Weights | trade_weight_ + quality, valuation, fair_value, analyst_upside, momentum, drawdown, dividend, group_consensus (0–100) |
| Taxes | pairing_mode (fifo, optimize, optimizeYearly); time_test_years (0–10); crypto_time_test, crypto_time_test_from, crypto_small_sale_from; deduct_fees; fx_method (market, cnb, unified); small_sale_limit / exempt_gains_cap, with _currency and _czk variants; tax_rate_pct (0–60) |
| Alerts | Rules, each with an optional scope (all, a class, category, sector or ticker; fair-value rules: all, a class or a ticker): alert_move_rule (days 1, 7, 30, 90, 180, 365); alert_drawdown_rule, alert_rebound_rule (days 30, 90, 180, 365, 730, 0 = all-time); alert_level_rule (optional metric: marketCap, trailingPE, forwardPE, evToEbitda, fcfYield, fcfPaybackYears, peg, pos52w, portfolioWeight); alert_fair_value_rule (model: weighted, dcf, graham, lynch, evEbitda, ddm, multiples, mine; cond: below, above, within); alert_events (7, 14, 30, 60 days; kinds: earnings, dividends, macro). 0 removes. |
| Data, Social | source_enabled (with source); social_enabled, social_presence, social_share_ + holdings, performance, allocation, risk, buy_plan, targets, plan_tier, platform, brokers, trade_events |
One-click reports ✨
Report cards sit below the assistant’s opening message. A click posts the report’s exact prompt as your message (with a show full prompt expander) and streams the report into the chat, where follow-ups build on it; clicking again regenerates it, and the card’s timestamp jumps back to the last one. Hover a card for what it contains. While a configured model cannot answer, the cards are dimmed under the reason line; an empty portfolio shows a single Reports need your portfolio card.
| Report | What you get |
|---|---|
| Portfolio brief | The state of the portfolio, today’s moves tied to actual headlines, events ahead, major risks and watch items. |
| Strategy check | Your portfolio against your saved thesis, horizon and risk tolerance: an alignment score, what fits, what diverges, sized moves to consider. Needs a saved strategy (⚙ Settings → Strategy). |
| Buy advice | Explains the current buy and sell orders, amounts and limits from ⚙ Settings → Trades, and why fewer orders or unspent money were planned. Needs an amount; cached until the plan, strategy or month changes. |
| Earnings season digest | Appears only as an earnings season wraps — once three holdings (or a quarter of a smaller portfolio) have reported within about six weeks: beat/miss verdicts, guidance changes, biggest surprises first, weighed by position size, with links to the filings. The Events card’s season line opens it too. |
Scheduled AI briefings
A scheduled AI briefing answers your own question on a timetable — “every Monday: what moved last week and why” — in the Insights & alerts bell and optionally by email. Set them up under ⚙ Settings → Alerts → AI briefings: a title, your request or one of four templates, daily, weekly or monthly at your hour, optionally a skill. Each run is the assistant alone with the same context and read-only tools, on your own model or else your hosted AI allowance; it describes rather than recommends trades. Open in chat continues a briefing in the conversation.
Proactive tips
When connected, the assistant can comment as you work — star a stock and it may flag your concentration in that sector. Set it with the Off / Minimal / Active slider under Proactive comments in ⚙ Settings → Assistant (default Minimal: only high-signal remarks). Tips use the fast model and a compact context, fire about 1.5 s after your action and at most once every 15 s, and wait while your own question is being answered. Loading the demo portfolio sets off no tip; the first thing you do with it can.
A tip has a faint green wash and opens with a bold line, written by the app, saying why it appeared — “Tip - You looked at AAPL:”, “Tip - You opened the Trades tab:”. The arrow beside Copy folds and opens it. Only the newest two tips stay open; a tip you reopen counts as new, one you fold stays folded, across reloads.
Investing notes 📝
Short, timestamped reminders you write for the assistant — “sell X before earnings”. Edit them behind the 📝 pill next to the asset search (every tab; shortcut N) or under ⚙ Settings → Strategy: up to 20 notes of 500 characters, newest first. Or ask: “add a note: watch my cash level this month”.
Notes travel in full with every chat message, whatever the Assistant context boxes say and however long the conversation grows, and with every tip and ✨ report. The assistant reads them last before your question, newest first, and where two disagree the newer wins. They are your standing intentions: an answer about buying, selling or holding addresses every note that bears on it, pushes back when your data disagrees, and never executes one on its own. Buy advice weighs them, and Trades-tab candidates a note mentions get a 📝 hint.
Assistant memory 🧠
The assistant keeps its own notebook of concise memories across chats: after a non-trivial answer it may file a short line (“worst PEG: INTC 3.1; advised trim”), or you say “remember that I prefer dividend stocks”. Each is stamped with the date, the accounts loaded and your portfolio value then, read as history, never as today’s figures. Up to 50 memories of 300 characters; the oldest rolls out when full. They ride every message while the Memories context box is checked (the default). Review, edit or delete them under ⚙ Settings → Assistant → View memories; the Assistant memory switch beside it (on by default) stops saving and sending them without deleting any.
Skills — reusable custom instructions
Skills are Markdown files of standing instructions — output formats, focus areas, opening routines. Manage them under ⚙ Settings → Assistant → Skills: upload a .md file, click a name to edit, untick to keep it unapplied, ✕ to delete; changes apply immediately. Limits: 10 skills, 16 KB each, 40 KB of enabled text in total. A skill may name actions such as show_card or sort_table to perform when its condition is met, but can never override the safety and grounding rules or make the assistant give financial advice. Need inspiration? adds four ready-made skills: dividend-focus, value-discipline, risk-guardrails, opening-checkin.
Targeting surfaces with applies-to
The first line may name where a skill applies — applies-to: chat, tips (lenient: Applies To =, an optional --- fence, commas or spaces). The line is removed before the model sees the text. Without it, the skill applies to the chat only.
| Target | Where the skill applies |
|---|---|
chat | Every chat message (the default). |
tips | Proactive tips. |
reports | The one-click ✨ reports: Portfolio brief, Strategy check, Buy advice and the Earnings season digest. |
appStart | The first exchange of a fresh chat session (clearing the conversation starts a fresh one). |
Example skills
An output format for the Portfolio brief:
applies-to: reports
When generating the Portfolio brief, open with a single-line verdict
(calm / watch / act), put every number in bold, and close with
"Next review:" and a suggested date.
A tips focus:
applies-to: tips
Include the concrete amount at stake whenever you mention a position,
and prefer tips about concentration risk over daily price moves.
An app-start routine that uses actions by name:
applies-to: appStart
On app start, use the show_card action with card "allocation" and comment
on the two largest weights.
Your own agent over MCP
The ChatGPT app, the Claude app, Claude Code, Codex or any other MCP client can read your portfolio with its own model and, with your approval, act on it. Connecting your own portfolio needs Fortunest Pro or the 14-day Pro trial; a free demo endpoint needs no account at all. Setup for every client, what an agent may do and troubleshooting are in the Fortunest MCP server chapter — start with the demo portfolio.
Where else the assistant shows up
- ✨ AI insights — up to four model-found patterns in the topbar Insights & alerts bell (Insights).
- ✨ Smart search — a quoted prompt in the topbar finder or in Explore (Finder).
- ✨ Semantic holdings filter — quoted text in the holdings table’s filter box.
- ✨ News & filing summaries — on asset detail panels and the Events card’s filed reports.
- Scheduled AI briefings — in the bell and by email (above).
The ✨ features use the fast model and stay disabled, with a red “not connected” note, until you connect.
News summaries are shared, not per-person
The ✨ summary above a stock’s headlines is not written for you alone: on the hosted app one summary per set of headlines is served to every reader. It appears instantly and costs your free AI allowance nothing unless you are first to ask about fresh headlines; the demo holdings and the largest listed companies (top 20) are refreshed daily. A shared summary names its model, gpt-5.6-luna.
Only the app’s own model writes shared summaries, and only finished ones are shared. A summary from your own provider — your OpenAI, Anthropic or OpenRouter key, or your Ollama or LM Studio server — is shown to you only; Refresh asks your model without replacing the shared one. The pool holds public article text and a ticker, with no account, session or holding; filing summaries, insights, reports and the chat stay yours.