AI Setup: Connecting an LLM Provider to SabrTrader
SabrTrader's AI features — AI Chat, Chart Pattern recognition and AI-generated alerts — are powered by a large language model that you supply through your own paid API key. This page documents the one-time configuration in Settings > Preferences, every setting on the Preferences page, how to choose a provider and model, where API keys come from, and how to control token usage cost. It covers configuration only; using the AI for analysis is documented separately.
Overview: what SabrTrader's AI integration is and is not
SabrTrader's AI integration is a bring-your-own-key architecture. The platform builds the prompts, gathers chart context and renders the results inside its AI panels, but the reasoning itself is performed by a third-party large language model (LLM) that you connect with your own API key. SabrTrader does not resell AI capacity and does not bill you for AI usage.
What this means in practice:
- You need a paid API account with one supported AI provider (Anthropic, OpenAI, xAI/Grok, Perplexity and other major providers).
- Every AI request is billed by that provider to your account, based on token usage — the amount of text sent and generated.
- Setup is a one-time job:
Settings > Preferences > AI Model. Select provider, select model, paste API key, press Test Connection. - Once configured, the AI becomes available everywhere it is exposed in the platform: the AI Chat panel and Chart Pattern panel inside Chart Trader on any chart, and AI-assisted features such as AI-generated alerts and AI strategy generation.
What this page does not cover: how to write effective prompts, how to interpret AI market-structure commentary, or how to build alerts and strategies with the AI. Those are separate topics. This page is strictly about getting the connection working correctly and keeping its cost predictable.
What the AI integration is not: it is not a signal service, not an autotrader, and not a substitute for reading the tape. AI output is written analysis and, optionally, drawn levels. Confirmation still comes from order flow tools such as the Footprint chart and the Liquidity Heatmap and SuperDOM.
Prerequisites
Before opening the AI Model settings page, confirm all of the following. Missing any one of them is the most common reason Test Connection fails.
| Requirement | Detail |
|---|---|
| Paid API account | An account on a supported provider's developer console with API access enabled. A consumer chat subscription (for example a $20/month ChatGPT Plus or Claude Pro plan) is a separate product and does not grant usable API access. |
| Funded credit balance | Most providers require a prepaid credit balance or an attached payment method before API requests are accepted. An unfunded key authenticates in some cases but returns billing errors on the first real request. |
| An API key you can copy | Keys are typically displayed exactly once at creation time. If you did not copy it, create a new one rather than hunting for the old value. |
| Model access on your tier | Some flagship models are gated behind spend tiers or organisation verification. A model can appear in a dropdown and still be unavailable to your specific account. |
| Cost awareness | Every AI request in SabrTrader consumes tokens on your provider account. Heavier use — long chats, large chart context, flagship models — costs more. |
Nothing about the AI setup affects your broker or data connection. You can configure the AI with only the default delayed dxFeed data connection active, before ever linking a broker such as Rithmic or Interactive Brokers.
Getting to the settings: Settings > Preferences
The AI configuration lives inside the platform-wide Preferences panel, not on an individual chart.
- Open the Settings menu from the SabrTrader Control Center or main menu bar.
- Choose Preferences.
- The Preferences panel opens with a left-hand list of categories. Each entry loads a different settings page into the right-hand pane.
- AI Model is one of those categories, listed alongside chart rendering, time zone, privacy, theme colour, language and MCP server.
Because you are already in Preferences, it is efficient to work down the whole list once and settle the general platform settings before opening AI Model. The next section documents every item.
Preferences reference: every setting on the page
The Preferences panel holds platform-wide behaviour. Settings here apply to SabrTrader as a whole rather than to a single chart or workspace.
Chart rendering
Chart rendering selects the graphics mode SabrTrader uses to draw charts (for example hardware-accelerated versus software rendering). The default is the recommended value for the large majority of machines. Inline help text next to the setting explains the circumstances in which a change is appropriate — typically visual artefacts, blank chart areas, or poor frame rates on older or virtualised graphics hardware. If charts look correct and scroll smoothly, leave this alone.
Time zone
Time zone sets the clock used for chart time axes, session boundaries and timestamps. The default is recommended. Most US futures and equities traders set this to US Eastern so that the 09:30 ET RTH open and 16:00 ET close land where they expect them, and so that overnight/ETH session boundaries match published exchange hours. Changing the time zone shifts every session-based tool — including session volume profiles — so pick one and keep it.
Privacy: hide accounts
The Privacy section contains a hide accounts option that masks account numbers, and account balances, everywhere they appear in the interface. Enable it before screen sharing, recording a video, streaming, or posting screenshots. It is a display mask only: it does not change routing, does not disconnect anything, and does not affect order entry.
Theme colour
Theme colour offers four built-in presets as starting points. Every colour in the scheme can then be customised individually — background, grid, up/down candles, text, panel chrome. Customisations are not automatically permanent: save the result as a template (a chart/workspace template) so the scheme can be reapplied to new charts and restored after an update or reinstall.
Language
Language sets the interface language for menus, labels and dialogs. It does not translate AI responses; response language is driven by the language of your prompt and the model itself.
MCP server
The MCP server page configures SabrTrader's Model Context Protocol interface. MCP is an open interface that allows an external AI tool or a developer's own code to connect to platform data and actions programmatically — for example, an external assistant querying chart data or issuing platform commands. This is a developer and integration feature. For normal trading, leave it disabled and untouched; it is not required for AI Chat or Chart Pattern to work.
AI Model
The AI Model page holds the provider selector, model selector, API key field and Test Connection button. It is documented in full in the next section.
AI Model settings reference
The AI Model page has exactly four controls. There is no advanced configuration, no endpoint URL to enter and no prompt template to edit.
| Control | Type | What it does |
|---|---|---|
| Provider | Dropdown | Selects which company hosts the model. Options cover the major AI providers: Anthropic (Claude), OpenAI, xAI (Grok), Perplexity and others. Your choice determines which API key format is valid and which models appear in the model list. |
| Model | Dropdown | Selects the specific model within that provider — for example Claude Opus-class or Sonnet-class models. The list is repopulated whenever the provider changes. The chosen model drives both cost per request and response latency. |
| API key | Text field | Where you paste the credential from the provider's console. It is paste-only in practice: keys are long random strings and typing them invites errors. No other credential is required. |
| Test Connection | Button | Sends a minimal request to the selected provider using the selected model and the entered key, and reports success or failure. Use it every time you change any of the three fields above. |
How the provider and model selectors interact
Provider and model are dependent fields. Changing the provider clears or replaces the model list, and a key issued by one provider will never authenticate against another. If you switch from Anthropic to OpenAI you must paste an OpenAI key; the previous Anthropic key is not reusable.
What Test Connection proves and what it does not
Test Connection confirms four things: the key is syntactically valid, the provider accepts it, the selected model is reachable by that key, and network egress from your machine to the provider is working. It does not guarantee your credit balance is large enough for sustained use, and it does not validate rate limits under load. A green result plus an exhausted balance still produces failures inside AI Chat later.
Concepts and definitions used on this page
Large language model (LLM)
A large language model is an AI model that interprets text prompts and supplied context and produces written output. In SabrTrader, the context includes information about the chart you are looking at — instrument, timeframe, recent price data — and the output is written analysis, sometimes accompanied by machine-readable price levels the platform can draw.
AI provider
An AI provider is the company that hosts and serves the model over an API. Anthropic serves Claude models, OpenAI serves GPT-class models, xAI serves Grok, Perplexity serves its own search-augmented models. The provider owns the billing relationship with you.
API key
An API key is a private credential string issued by your provider that authenticates requests and identifies the account those requests are billed to. It functions as a password. Anyone holding your key can spend your credit balance, so it must never be shared, screenshotted, committed to code or pasted into a public chat.
Token usage cost
Providers bill per token — a fragment of text, roughly three-quarters of a word on average. Both the text you send (input tokens, including the chart context SabrTrader attaches) and the text the model generates (output tokens) are billed, usually at different rates. Consequences: long conversations cost more than short ones because prior turns are resent; large context costs more than small; verbose answers cost more than concise ones; and flagship models cost several times more per token than mid-tier models.
Paid versus free model tiers
A consumer chat subscription buys access to a provider's own web or mobile chat app. An API plan buys programmatic access for third-party software. They are separate products with separate billing. SabrTrader requires the latter. Free tiers generally expose either no API access at all or a trial quota too small and too rate-limited to be usable.
Auto top-up (auto add funds)
Auto top-up is a provider billing option that automatically recharges your prepaid credit balance — commonly by $20, $50 or $100 — whenever it falls to zero or below a threshold. It prevents interruptions but removes your natural spending brake, which is how users end a month surprised by the total. SabrTrader's recommendation is to disable it and top up manually.
Model selection trade-off
The model selection trade-off is the choice between flagship models (most capable multi-step reasoning, highest price per token, slowest) and lower-cost models (fast, cheap, and frequently indistinguishable in quality on routine chart questions). Matching the model to the difficulty of the task is the single largest lever on your AI bill.
MCP server (Model Context Protocol)
The Model Context Protocol is a standard interface that lets external AI tools and developer code connect to an application's data and actions. SabrTrader's MCP server setting exposes that interface for integration work. It is unrelated to, and not required by, the built-in AI panels.
Market structure
Market structure is the sequence of swing highs, swing lows and ranges that determines whether price is trending (higher highs and higher lows, or lower highs and lower lows) or balancing inside a range. "What's the market structure right now?" is the canonical first prompt for AI Chat. SabrTrader also has a dedicated Market Structure indicator that computes this mechanically.
Support and resistance
Support and resistance are price areas where previous trade has repeatedly halted or reversed movement — support below current price, resistance above. The AI can identify candidate levels from the visible chart and offer them for one-click drawing.
Timeframe context
Timeframe context is the chart interval the AI analyses, such as five minutes. It determines how much detail and how much history the answer reflects. A five-minute chart answer describes intraday structure; a daily chart answer describes multi-week structure. The same prompt on two timeframes legitimately produces two different answers.
Prompt
A prompt is the question or instruction you type into AI Chat. It defines the scope of the response. Specific prompts that name the instrument, timeframe and desired output produce usable answers; vague prompts produce generic commentary. See prompt engineering for AI alerts for the same principle applied to alert creation.
Chart pattern recognition
Chart pattern recognition is AI-assisted detection of recurring formations on the chart. In SabrTrader it is a separate panel from AI Chat, reachable from the same Chart Trader menu, because it runs a structured detection task rather than an open-ended conversation.
Chart Trader
Chart Trader is the panel docked to a SabrTrader chart for order entry, trade management and chart-side tools. The AI Chat and Chart Pattern entries live inside it. Chart Trader is also the surface from which orders are placed and, when the Trade Copier is armed, copied.
Template
A template is a saved set of chart, theme and layout settings you can reload on demand. Saving a customised colour scheme as a template is what makes it survive beyond the current chart.
Choosing a provider and a model
There is no single correct answer. All major providers produce competent chart commentary. Provider choice is largely personal preference, driven by which account you already pay for and how consistent you find the output style.
Provider selection guidance
| Provider | Typical reason to choose it |
|---|---|
| Anthropic (Claude) | Consistency and predictability of output format; the provider used in the demonstration for exactly that reason. Useful when you want the same answer shape every time. |
| OpenAI | Broadest model range and price ladder; a natural choice if you already hold an OpenAI developer account. |
| xAI (Grok) | Alternative frontier models; choose if you already have API credit there. |
| Perplexity | Search-augmented answers; useful when you want commentary that can reference outside information rather than the chart alone. |
| Other supported providers | Listed in the provider dropdown. Any of them work identically from SabrTrader's side once a valid key is entered. |
Model tier selection guidance
| Task | Suggested tier | Reasoning |
|---|---|---|
| "Give me support and resistance on a 5-minute chart" | Mid or low cost | A bounded, single-step extraction task. A flagship model is overkill and costs several times more for no practical gain. |
| "What's the market structure right now?" | Mid cost | Short structural summary; mid-tier models handle this reliably. |
| Chart Pattern scans | Mid cost | Structured detection with a defined output; rarely benefits from frontier reasoning. |
| Multi-step reasoning: combining session profile levels, delta behaviour and a trade plan with conditions | Flagship (Opus-class / frontier) | Long chains of dependent reasoning are where frontier models measurably outperform. |
| Generating complex alert or strategy logic | Flagship | Errors are expensive to debug; higher capability pays for itself here. |
Recommended approach: start on a mid-tier model as your daily driver. Escalate to a flagship model only for a specific complex question, then switch back. Because changing the model is a two-click operation on the same settings page, escalating on demand costs nothing but a few seconds.
Model names change frequently — an Opus 4.8 today may be superseded by an Opus 5 tomorrow. Treat the dropdown as the source of truth for what is currently available, not any list published elsewhere.
Where to get an API key
Each provider issues keys from its own developer console, and each console is laid out differently and changes over time. Rather than following a screenshot that goes stale, use the shortcut: ask that provider's own chat assistant how to create an API key for your account. It will return current, provider-specific instructions.
Generic procedure (applies to every provider)
- Sign in to the provider's developer console or platform site — this is usually a different site from the consumer chat app, though it may accept the same login.
- Confirm or set up billing: add a payment method or purchase prepaid credits. Many providers grant a small starter credit (commonly around $20) but block API calls until billing exists.
- Open the API keys section (sometimes under Settings, Organisation or Account).
- Choose Create key / New secret key. Give it a recognisable name such as
SabrTraderso you can revoke it selectively later. - Copy the key immediately. Most providers display the full value exactly once and only show a truncated fingerprint afterwards.
- Paste it into
Settings > Preferences > AI Model > API keyin SabrTrader. - Press Test Connection.
- Store the key somewhere private if you need it again — a password manager, not a text file on the desktop.
Key hygiene
- Use a separate key per application. If a key leaks you can revoke just that one.
- Revoke keys you no longer use. Revocation is immediate and irreversible.
- Never paste a key into a public chat, forum post, screenshot or video. Enable the Preferences privacy option and blur the key field before recording.
- After rotating a key (revoking the old one and creating a new one), paste the new value into SabrTrader and press Test Connection again — the old key stops working the moment it is revoked.
Managing usage and cost
Because billing sits with the provider, cost control also sits with the provider. SabrTrader's role is to let you pick a cheaper or more expensive model; everything else is done in the provider console.
How provider billing typically works
- You hold a prepaid credit balance. Requests draw it down according to token usage.
- New accounts often receive a starter credit — commonly around $20 — which is easy to confuse with a consumer subscription of the same price. They are different things.
- When the balance reaches zero, API requests fail. In SabrTrader this appears as AI Chat returning an error or stopping mid-answer.
Auto top-up: turn it off
Providers offer auto top-up (also labelled auto add funds, auto-recharge or auto top-off): when the balance hits zero it silently charges your card another $20, $50 or $100 and continues. If you use the AI heavily this can repeat many times in a month with no interruption to warn you.
Recommendation: disable auto top-up and top up manually in $50–$100 increments. When the balance runs out you notice, you decide whether the spend was justified, and you refill deliberately. This converts an invisible recurring charge into a conscious decision.
Additional cost controls
- Set a hard monthly spend limit in the provider console where one is offered. This is a ceiling that cannot be exceeded, unlike a soft alert.
- Configure usage alert emails at, for example, 50% and 80% of your intended monthly spend.
- Check the provider's usage dashboard weekly at first, so you learn what a typical session costs you in practice.
- Keep a lower-cost model selected by default and switch up only when needed.
- Ask fewer, better prompts. One specific question costs less than five vague follow-ups, because each follow-up resends the conversation so far.
- Start a new chat when you change topic, rather than extending a long thread — long threads carry an ever-growing input token cost.
Tutorial: complete AI setup, step by step
- Open the settings. Go to Settings > Preferences. The Preferences panel opens with the category list on the left.
- Set chart rendering. Leave it at the default/recommended value. Read the inline guidance text and change it only if you are experiencing rendering artefacts or poor performance.
- Set the time zone. Choose the zone you want chart timestamps and session boundaries expressed in. Most US traders select US Eastern.
- Set privacy. Enable hide accounts if you will be sharing your screen, recording or streaming.
- Set theme colour. Pick one of the four presets, then customise individual colours if you wish. If you customise, save the result as a template so it is not lost.
- Set language. Choose your interface language.
- Skip MCP server unless you are doing developer integration work. Leave it as-is for normal trading.
- Open the AI Model page from the category list.
- Select a provider from the provider dropdown (for example Anthropic Claude, OpenAI, Grok or Perplexity). Choose the provider where you hold — or intend to open — a paid API account.
- Select a model from the model dropdown, which repopulates for the chosen provider. Start with a mid-tier model unless you specifically need frontier reasoning.
- Paste the API key into the API key field. Copy and paste it; do not retype. Verify there is no leading or trailing whitespace and that the whole string was captured.
- Press Test Connection. Wait for the result. A success message means provider, model and key agree and the network path works. A failure means one of those three is wrong — see Troubleshooting.
- Save and close the Preferences panel.
- Open a chart on any instrument and timeframe — a 5-minute chart is a good first test because it produces enough structure to discuss.
- Expand Chart Trader on that chart and choose AI Chat from its tool list. Note the separate Chart Pattern entry alongside it.
-
Send a prompt. Either click one of the built-in example prompts or type your own, such as
What's the market structure right now? - Wait for the response. Generation takes a few seconds and longer on flagship models or with large context. The panel indicates that it is working.
- Review the answer. Read it as context, not instruction. If the response includes candidate levels, an Apply control appears.
- Optionally apply the levels. Pressing Apply draws the AI's suggested support and resistance levels directly onto the chart as objects you can then move, edit or delete like any manual drawing.
Total time for a first-time setup, assuming the provider account already exists: under two minutes. Creating the provider account and funding it usually takes longer than configuring SabrTrader.
Verifying the setup on a chart
A successful Test Connection is a technical check. The functional check is asking the AI a real question on a real chart.
Chart Trader and the AI entries
Chart Trader is the panel docked to the chart. Expanding it reveals order-entry controls plus the chart-side tool list. Two AI entries live there:
- AI Chat — an open conversation panel. You type any question; the model answers using the chart's instrument, timeframe and recent data as context.
- Chart Pattern — a dedicated pattern-recognition panel that scans the chart for recurring formations and reports what it detects. It is separate from AI Chat because it performs a structured detection task rather than free-form conversation.
Example prompt presets
AI Chat ships with a small set of example prompts. They exist purely to get you started; they are not a menu of the only supported questions. Any prompt is allowed. Useful starting prompts include:
What's the market structure right now?Give me the key support and resistance levels on this 5-minute chart.Is price trending or balancing, and what would invalidate that read?Where is the most recent swing high and swing low?
Response behaviour
- Responses take a few seconds to generate. Flagship models and longer context take longer.
- Answers stream in progressively rather than appearing all at once.
- When the model proposes price levels, SabrTrader surfaces an Apply action. One click draws them on the chart.
- Applied levels are ordinary chart objects: editable, movable and deletable. Applying does not place orders and does not modify any position.
If AI Chat returns a coherent answer, the integration is fully working. From here, the productive next steps are learning to use it — see AI-generated alerts and Auto Strategy Builder for AI features that build on the same connection.
Best practices
- Press Test Connection immediately after pasting a key. Finding out the key is wrong during a live session is far worse than finding out in Preferences.
- Keep a lower-cost model as your daily driver. Reserve flagship models for genuinely multi-step questions. This typically cuts AI spend by a large multiple with no loss of usefulness on routine queries.
- Disable auto top-up and top up manually in $50–$100 increments. Add a hard spend limit in the provider console where available.
- Enable the privacy / hide accounts option before recording or streaming, and blur or clear the API key field before any screen capture.
- Save your theme and layout as a template before experimenting. Customised colour schemes are not preserved automatically; a template is the only reliable way to recover one.
- Use a dedicated, named API key for SabrTrader so you can revoke it independently of other tools.
- Re-test the connection after rotating keys or changing provider, and after any provider announcement about model deprecation.
- State the timeframe in your prompt. Even though the chart supplies context, naming the interval explicitly produces tighter answers.
- Treat AI output as context, not a signal. Confirm any level or structural read with order flow — footprint, delta and the liquidity heatmap — before acting.
- Practise AI-assisted reads in Market Replay so you can test whether the AI's commentary would have helped, without risking capital.
- Start fresh chats for new topics to avoid paying to resend an irrelevant conversation history.
- Check the provider usage dashboard weekly for the first month until you know your own cost baseline.
Common mistakes
| Mistake | Why it fails | Fix |
|---|---|---|
| Trying to use a free or consumer chat subscription | A ChatGPT Plus or Claude Pro plan grants access to the provider's own chat app, not API access. There is no key to paste. | Open a paid API/developer account with the provider and fund it. The two products are billed separately. |
| Pasting the wrong credential | Session tokens, organisation IDs, project IDs and workspace IDs are not API keys. They will not authenticate. | Copy the value from the provider's API keys page specifically, created via Create key. |
| Truncated paste | Double-click selection or a manual drag frequently misses the first or last characters of a long key. | Use the console's copy button, then Test Connection to confirm. |
| Leaving a flagship model selected for trivial queries | You pay frontier-tier token rates for a question a mid-tier model answers identically. | Set a mid-tier default; escalate manually for hard questions. |
| Leaving auto top-up enabled | The balance refills silently and repeatedly, producing a surprising month-end total. | Disable auto top-up; top up manually; set a hard spend limit. |
| Not saving a customised colour scheme as a template | Custom colours are working state, not a persisted preset. They can be lost on reset, reinstall or when opening a new chart. | Save the scheme as a template immediately after building it. |
| Assuming SabrTrader bills for AI usage | SabrTrader never charges for tokens. All AI cost is billed by the provider to your own account, so watching only your SabrTrader invoice hides the real spend. | Monitor the provider's usage dashboard, not the platform. |
| Expecting AI output to be a trade signal | The model produces written analysis from chart context. It has no risk model, no position awareness and no execution edge. | Use AI output as one input. Confirm with order flow and your own plan before acting. |
| Changing provider but keeping the old key | Keys are provider-specific. An Anthropic key will never authenticate against OpenAI. | Paste a key from the newly selected provider and re-test. |
| Enabling and configuring the MCP server "just in case" | MCP is a developer integration surface, unrelated to AI Chat. Changing it adds risk and complexity with no trading benefit. | Leave MCP server alone unless you are building an integration. |
| Recording a walkthrough with accounts and key visible | Account numbers and API keys are sensitive credentials; publishing them exposes your account. | Enable Privacy > hide accounts and clear the key field before capturing anything. |
Frequently Asked Questions
Does SabrTrader include AI usage, or do I need my own API key?
You need your own API key. SabrTrader is bring-your-own-key: the platform supplies the AI features — AI Chat, Chart Pattern, AI-generated alerts and AI strategy generation — but the model runs at a third-party AI provider that bills you directly for token usage. SabrTrader does not charge for AI requests and does not resell AI capacity. Configure the key once in Settings > Preferences > AI Model.
Which AI providers can I connect to SabrTrader?
The provider dropdown on the AI Model page lists the major providers, including Anthropic (Claude), OpenAI, xAI (Grok) and Perplexity, plus other supported providers. Any of them work identically from SabrTrader's side once a valid key for that provider is entered. Provider choice is largely personal preference and depends on where you already hold API credit. Because the dropdown is updated with the platform, treat it as the authoritative list.
Can I use a free ChatGPT or Claude plan for SabrTrader's AI features?
No. Free tiers, and even paid consumer chat subscriptions such as ChatGPT Plus or Claude Pro, grant access to the provider's own chat app — not programmatic API access. There is no API key to paste. You must open a paid API/developer account with the provider and fund a credit balance. The consumer subscription and the API plan are separate products with separate billing, so paying for one does not include the other.
Which model should I choose for chart analysis?
Start on a mid-tier model. Routine tasks — "give me support and resistance on a five-minute chart", "what's the market structure right now?" — are bounded single-step questions that mid and low-cost models handle as well as flagship models, at a fraction of the token cost. Reserve flagship (Opus-class / frontier) models for genuinely multi-step reasoning: combining profile levels with delta behaviour, or generating complex alert and strategy logic. Switching models takes two clicks on the same settings page, so escalate on demand rather than paying for frontier capability permanently.
How much does SabrTrader's AI cost to run?
It depends entirely on your provider, your chosen model and how heavily you use it, because billing is per token — the amount of text sent and generated. A light user asking a handful of short questions per session on a mid-tier model spends very little. A heavy user running long conversations on a flagship model can spend many times more. Practical approach: fund $50, use the AI as you normally would for a week, then read the provider's usage dashboard. That gives you a real baseline rather than an estimate.
Where do I find my provider's API key?
In that provider's own developer console, under an API keys section, via a Create key action. Every console is laid out differently and changes over time, so the fastest reliable route is to ask that provider's own chat assistant how to create an API key for your account — it will give current, provider-specific steps. Generic flow: sign in to the developer console, set up billing, open API keys, create a key named SabrTrader, and copy it immediately because most providers display the full value only once.
Should I enable auto top-up on my AI provider account?
SabrTrader's recommendation is no. Auto top-up (auto add funds / auto-recharge) silently charges your card another $20, $50 or $100 each time your credit balance hits zero. With heavy AI use this can repeat many times a month with nothing to alert you, producing an unpleasant surprise on the statement. Disable it and top up manually in $50–$100 increments so each refill is a conscious decision, and set a hard monthly spend limit in the provider console if one is offered.
What does the Test Connection button actually check?
It sends a minimal request to the selected provider using the selected model and the entered key. A success result confirms four things: the key is valid, the provider accepts it, the selected model is reachable by that key, and your machine can reach the provider over the network. It does not confirm that your credit balance is sufficient for sustained use, and it does not test rate limits under load — so a passing test can still be followed by billing or throttling errors inside AI Chat later.
Where is AI Chat located once the AI is set up?
On any chart, inside the Chart Trader panel. Expand Chart Trader and choose AI Chat from its tool list. The Chart Pattern panel sits alongside it in the same menu. AI Chat is chart-scoped: it uses the instrument and timeframe of the chart it was opened from as context, so opening it on a five-minute chart produces intraday answers and opening it on a daily chart produces longer-horizon answers.
What is the MCP server setting for?
MCP stands for Model Context Protocol — a standard interface that lets external AI tools or developer code connect to an application's data and actions programmatically. SabrTrader's MCP server setting exposes that interface for developers building integrations against the platform. It is not needed for AI Chat, Chart Pattern or any built-in AI feature. For normal trading, leave it alone.
Can I change the AI model later without redoing setup?
Yes. Return to Settings > Preferences > AI Model and pick a different model from the dropdown. The API key stays as-is provided you remain on the same provider, and the change takes effect for subsequent AI requests. If you change the provider rather than just the model, you must paste a key issued by that new provider — keys are provider-specific — and press Test Connection again.
Is my API key stored securely, and is chart data sent to the provider?
The key is stored locally as part of your platform settings and is used only to authenticate requests to the provider you selected. Treat it as a password: do not share it, do not include it in screenshots or recordings, and revoke it at the provider if it is ever exposed. When you send a prompt, SabrTrader transmits your question plus the relevant chart context (instrument, timeframe, recent price data) to that provider so the model can answer. If you are subject to policies restricting what market or account data may leave your machine, review your provider's data-retention terms before enabling the AI features.
How do I hide my account numbers before recording or streaming?
Open Settings > Preferences and enable the Privacy > hide accounts option. It masks account numbers and balances everywhere they appear in the interface. It is purely a display mask: connections stay live, orders route normally and nothing about your trading changes. Enable it before you start capturing, and also clear or blur the API key field on the AI Model page, since a key is as sensitive as an account number.
How do I keep a customised theme colour scheme?
Save it as a template. The Preferences theme section offers four presets as starting points, and you can then customise every colour individually — but those customisations are working state, not a persisted preset. Saving the result as a chart/workspace template lets you reapply the scheme to new charts and restore it after a reset, reinstall or update. Save the template as soon as you are happy with the colours, before experimenting further.
What's the difference between AI Chat and the Chart Pattern panel?
AI Chat is an open-ended conversation: you type any prompt and receive written analysis, optionally with levels you can apply to the chart. Chart Pattern is a structured task: it scans the chart for recurring formations and reports what it detects, without requiring you to phrase a question. Both are reached from the same Chart Trader tool list and both consume tokens on the same API key you configured on the AI Model page.
Do I need a broker connection before configuring the AI?
No. The AI configuration is independent of market data and order routing. You can select a provider, paste a key and pass Test Connection with only the default delayed dxFeed data connection active. To get useful AI answers you do need a chart with data on it, which the default feed provides. Live broker connections such as Rithmic, Schwab or Interactive Brokers can be added at any time before or after.
Can the AI place trades for me?
No. The AI produces written analysis and, when relevant, candidate price levels you can draw on the chart with one click. Applying levels creates chart objects only — it never submits, modifies or cancels an order, and never changes a position. All order entry remains manual through Chart Trader or the SuperDOM, or automated through a strategy you explicitly deploy via AlgoStudio Pro.
Why does the same prompt give different answers on different charts?
Because timeframe context is part of the input. AI Chat analyses the chart it was opened from, so its answer reflects that instrument's interval and visible history. "What's the market structure right now?" on a five-minute chart describes intraday swings; the same prompt on a daily chart describes multi-week structure. Neither answer is wrong. Name the timeframe explicitly in your prompt when you want to be certain which horizon is being described.
Does using a cheaper model make the AI's chart analysis unreliable?
Not for routine tasks. The tasks most traders use AI Chat for — summarising market structure, extracting support and resistance, describing whether price is trending or balancing — are bounded and single-step, and mid or low-cost models handle them consistently. The measurable advantage of flagship models appears in long chains of dependent reasoning: combining several data sources, generating complex conditional logic, or producing multi-part plans. Use the cheap model as the default and escalate for those specific cases.
What happens if my provider credit runs out mid-session?
API requests start failing. In SabrTrader this shows up as AI Chat returning an error, producing nothing, or stopping part-way through an answer. The platform itself, your charts, your data feed and your orders are unaffected — only the AI features stop. Fix it by topping up your balance in the provider console, then send a fresh prompt. If you disabled auto top-up as recommended, this interruption is the intended signal that you have reached your chosen spend for the period.
Troubleshooting
Test Connection fails
Work through these in order:
- Key completeness. Re-copy the key using the console's copy button and paste again. Partial selections are the most common cause. Check for leading/trailing spaces.
- Correct provider. Confirm the provider dropdown matches the provider that issued the key. Keys are never cross-compatible.
- Model availability. Some models are gated behind spend tiers or organisation verification. Select a widely available mid-tier model and re-test; if that succeeds, the problem is model access, not the key.
- Billing. Confirm a payment method or prepaid credit exists in the provider console. Unfunded accounts are frequently blocked at the API layer.
- Key status. Verify the key has not been revoked, deleted or rotated. Revocation is immediate — create a new key if in doubt.
- Network and region. Corporate firewalls, VPNs and region restrictions can block the provider endpoint. Test from an unrestricted network.
Wrong credential pasted (session token, project ID, organisation ID)
Only a value created on the provider's API keys page via Create key will authenticate. Session tokens from a browser, project IDs, organisation IDs and workspace identifiers look similar but are not credentials. Go back to the API keys page, create a new key named SabrTrader, copy it at creation time and paste that value.
AI Chat returns nothing, or errors part-way through an answer
Most likely causes, in order of frequency:
- Credit exhausted. Check the provider balance and top up.
- Rate limit hit. Too many requests in a short window. Wait a minute and retry, or reduce request frequency.
- Model deprecated or withdrawn. Providers retire models. Select a currently listed model and press Test Connection.
- Transient provider outage. Check the provider's status page before changing anything.
After any change, re-run Test Connection and then re-send the prompt.
Responses are very slow
Latency scales with model capability and context size. If answers take longer than you can tolerate:
- Switch from a flagship (Opus-class / frontier) model to a mid-tier model.
- Start a new chat instead of continuing a long thread — long threads resend the whole conversation as input.
- Ask a narrower question. Broad prompts generate long answers, and every generated token adds time.
A few seconds of generation time is normal and not a fault.
The model dropdown lists a model I cannot actually use
The dropdown reflects models the provider offers, not necessarily models your specific account is entitled to. Access to frontier tiers is often gated behind cumulative spend, organisation verification or a waitlist. Select a broadly available model, confirm with Test Connection, and check your provider console for the requirements to unlock the higher tier.
AI answers do not match the chart I am looking at
Confirm you opened AI Chat from the chart you mean to analyse — the panel is chart-scoped and inherits that chart's instrument and timeframe context. Also confirm the chart has loaded data; a chart with no history gives the model nothing to work with. State the instrument and interval explicitly in the prompt (for example "on this 5-minute MES chart") to remove ambiguity.
Charts render oddly: artefacts, blank areas or stuttering
Return to Settings > Preferences and revisit the Chart rendering setting. Read the inline guidance text next to it, which explains which alternative mode suits which situation, and change the mode only as that text advises. Restart SabrTrader after changing rendering mode. If the problem persists on every mode, the cause is likely graphics drivers or a virtualised display rather than the platform setting.
Account numbers or balances visible in a recording or screenshot
Enable Settings > Preferences > Privacy > hide accounts before capturing. It masks account identifiers and balances throughout the interface without affecting connections or order routing. Also clear the API key field on the AI Model page before capture — a key is as sensitive as an account number and cannot be masked by the privacy option.
A customised colour scheme disappeared
Custom theme colours are working state until you save them. Rebuild the scheme and immediately save it as a template. Templates persist and can be reapplied to new charts and restored after resets, reinstalls or updates. Keep one template as your known-good baseline before experimenting with new colours.
Unexpectedly large provider bill at month end
Two contributing causes are almost always present: auto top-up was enabled, so refills happened silently, and a flagship model was selected for routine queries. Fix both: disable auto top-up in the provider console, set a hard monthly spend limit and usage alerts, and change the SabrTrader model selector to a mid-tier model. Then review the provider usage dashboard weekly until you know your baseline.
AI worked yesterday but fails today after no changes on my side
Check, in order: provider credit balance (may have hit zero overnight), key validity (an administrator or a rotation policy may have revoked it), and model status (the model may have been deprecated). Press Test Connection to identify which layer is broken, switch to a currently listed model if needed, and re-test.
Glossary
- AI provider
- The company that hosts and serves the AI model over an API and holds the billing relationship with you. Supported providers include Anthropic (Claude), OpenAI, xAI (Grok) and Perplexity, among others.
- API key
- A private credential string issued by an AI provider that authenticates requests and identifies the account those requests are billed to. It functions as a password and must never be shared, screenshotted or published.
- API key revocation
- Invalidating a previously issued API key so it can no longer authenticate. Used when a key is unused, misplaced or compromised. Revocation is immediate, so SabrTrader must be given the replacement key and re-tested.
- Auto top-up / auto add funds
- A provider billing option that automatically recharges your prepaid credit balance whenever it is exhausted. It prevents interruptions but hides spending, so SabrTrader recommends disabling it and topping up manually in $50–$100 increments.
- Chart pattern recognition
- AI-assisted detection of recurring formations on a chart, exposed in SabrTrader as a Chart Pattern panel separate from AI Chat because it performs a structured detection task rather than open conversation.
- Chart rendering
- The graphics mode SabrTrader uses to draw charts. It should be left at the recommended default unless rendering artefacts, blank chart areas or poor performance indicate a compatibility problem; inline help text explains when to change it.
- Chart Trader
- The panel docked to a SabrTrader chart for order entry, trade management and chart-side tools. The AI Chat and Chart Pattern entries are located inside it.
- Large language model (LLM)
- An AI model that interprets text prompts and supplied context and produces written output. In SabrTrader an LLM receives the chart's instrument, timeframe and recent price data along with your prompt, and returns written analysis and optional price levels.
- Market structure
- The sequence of swing highs, swing lows and ranges that defines whether price is trending or balancing. Higher highs with higher lows indicate an uptrend; lower highs with lower lows indicate a downtrend; overlapping swings indicate balance.
- MCP server (Model Context Protocol)
- An interface that lets external AI tools or developer code connect to SabrTrader data and actions programmatically. It is a developer integration feature and is not required by the built-in AI Chat or Chart Pattern panels.
- Model selection trade-off
- The choice between flagship models (most capable at multi-step reasoning, highest price per token, slowest) and lower-cost models (fast, cheap, and often equally effective on routine chart questions). Matching model capability to task difficulty is the main lever on AI cost.
- Order flow
- The sequence of executed trades and changes in resting liquidity that drives price, studied directly through footprint, delta and DOM tools rather than inferred from price patterns. It is the recommended confirmation layer for any AI-suggested level.
- Paid vs free model tiers
- The distinction between a provider's consumer chat subscription and its paid API plan. SabrTrader's AI features require a paid API plan; free tiers and consumer chat subscriptions do not expose usable API access.
- Prompt
- The question or instruction typed into AI Chat, which determines the scope and usefulness of the response. Specific prompts naming instrument, timeframe and desired output produce usable answers; vague prompts produce generic commentary.
- Prompt engineering
- Writing a clear, specific instruction so the AI applies the exact indicators, session windows, levels and annotations you want, rather than inferring them.
- Support and resistance
- Price areas where previous trade has repeatedly halted or reversed movement — support below current price, resistance above. The AI can identify candidate levels from chart context and offer them for one-click drawing.
- Template
- A saved set of chart, theme and layout settings that can be reloaded on demand, used here to preserve a customised colour scheme so it survives resets, reinstalls and new charts.
- Timeframe context
- The chart interval the AI analyses, such as five minutes, which determines how much detail and history the answer reflects. The same prompt on different intervals legitimately produces different answers.
- Token usage cost
- Per-request billing charged by the AI provider based on the amount of text sent (input tokens, including attached chart context) and generated (output tokens). Longer conversations, larger context and more capable models all increase cost.
- Workspace
- A saved SabrTrader layout containing charts, DOMs and panels bound to specific connections, which can be reopened and switched.