The built-in assistant
The built-in assistant is a panel at the side of RigReady's window. It talks to the AI provider you chose in Settings (your own key from Anthropic, OpenAI, Google Gemini, OpenRouter or Mistral, or a model on your own PC through Ollama or LM Studio; see Choose who answers), and it works through the same tools an assistant connected over MCP gets, with the same safeguards.
Turn it on
Choose who answers in Configure > Settings, under AI, and store the key it needs (a server on your own PC needs none). The Assistant button then appears at the top of the window, beside Find or run. Until then there is no button; Open the assistant in the command palette opens the panel, says a provider is needed and takes you to Settings. What assistants connected over MCP drafted is shown in the panel either way. The binding guide has a link to the panel too.
The conversation fills the panel and the message box stays at the bottom. Drag the panel's left edge to make it wider or narrower (or use the arrow keys on it); RigReady remembers the width, and a double click on the edge brings back the usual one.
Every request is shown first
Type a message and press Prepare (or Enter). RigReady builds the request and shows it above the message box: the host and the model it goes to, which request of the question it is, about how many tokens it has and what reading it is estimated at, and what it carries, in a line each: your message, what the tools answered since the last request, how long the conversation is, and RigReady's instructions and tool list. Show the exact request shows the whole text. Nothing is sent until you press Send (or Enter again). Don't send throws the request away and leaves your message in the box; changing the message does the same, so the request on screen is always the one for the text in the box.
When the model uses a tool, its answer goes back with the next request, and that request is shown again before it goes. Tick Don't show this again for this conversation and the rest of that conversation goes without stopping. The tick belongs to that one conversation: New conversation starts over with the preview on again.
The preview names the one host the request goes to, for example api.anthropic.com, or the server on your PC. If you change the provider in Settings after a request was shown, that request is not sent: it is made again for the new settings and shown first. What a request never carries: your API key (it is the request's credential, sent in a header to that host and nowhere else) and files of yours other than what a tool answered. In what the tools answered, the path of your Windows user folder is written as {USER} however it is written there (in any case, in its short form such as C:\Users\JANESM~1, or in a file:/// address), and the PC's name as {PC} when it has five or more characters; other names and paths go as they are. The provider's terms and privacy policy apply to what is sent; with a model on your PC nothing leaves it. Exactly what is sent in Settings lists all of it.
While it works
A line above the message box says what the assistant is doing, for example "Waiting for the answer (request 2)" or "Running: Find installed programs (request 2)", with the tokens of the question so far and the estimate for them. Stop is beside it the whole time. Stopping loses nothing: a message that was not answered is back in the box, and what the tools already answered waits until you press Continue.
Every tool the assistant used is a row in the conversation: what it did in words, whether that read, changed something or launched a game, and a line about what it answered. Click the row to see the exact answer, which is what goes to the provider with the next request. A proposal or a script appears in the conversation at the place where it was made.
Replies are shown formatted: headings, lists, bold, code. An https:// link in a reply opens in your browser when you click it; any other address is shown as text. HTML in a reply is shown as text, and pictures are never loaded.
Estimates, prices and limits
Your provider counts the tokens of every request and reports them: those numbers are exact. The dollars are not. RigReady works them out from the token counts and from prices per million tokens kept in Settings, and those prices can be out of date. So every dollar figure RigReady shows is an estimate, and says which prices it was worked out from: "built-in prices of November 2026" (the prices that came with RigReady, with the month they were recorded) or "your own prices". None of them is a bill. Your provider's console shows what you are charged, and a spending limit set there is the one that holds: set one there.
Where prices are set for the model, the panel shows:
- Before a request goes, for example "Request 1 of this question · about 4,400 tokens · estimated $0.02 to read at built-in prices of November 2026, plus the answer". One question often takes several requests, one more each time the assistant uses tools.
- While it works, the tokens of the question so far and the estimate for them.
- Under the message box, the total of the conversation: requests, tokens, how many of them the provider read from its cache, and the estimate, with the line "Estimates, not a bill: your provider's console shows what you are charged."
Prices
Prices for this model, in dollars per million tokens, in Settings under AI, has four fields: input, output, cache read and cache write. For the Anthropic and OpenAI models RigReady lists, they start filled with its built-in prices, and the row says the month those were recorded. Providers change their prices. Compare the fields with your provider's price list, type your own numbers where they differ, and press Save prices: from then on the estimates for that model use your numbers and say "at your own prices". Reset to built-in goes back. Prices are kept for each model.
For any other model, and for a server on your own PC, the fields start empty. Until you type prices, RigReady shows no dollar figure for that model, and neither limit below applies to it. Type 0 where your provider charges nothing.
If the prices are changed after a request was shown, that request is not sent: it is made ready again with its new estimate, and shown first.
The ceiling for one question
Ask before the estimate for one question passes, in Settings under AI, is $0.25 unless you change it. Before a request would take the estimate for a question past it, RigReady stops, also when the preview is switched off for the conversation, says so in words that name the estimate, and shows that request with Continue past $0.25 in place of Send. Pressing it is good for as much again. Set the amount to 0 and RigReady never stops a question on its estimate.
The monthly budget
RigReady adds up the estimates of every request the assistant and the binding guide got an answer to in the calendar month, each at the prices in force when it was sent, and shows the sum in Settings. Stop sending when the estimate for a month reaches, in the same place, is a budget for that sum. It is off until you set an amount. With an amount set:
- From 80 percent of the amount, Settings says so, and so does every request that waits for Send.
- At the amount, or before a request whose estimate would take the month past it, RigReady sends nothing more. The message starts "Stopped on RigReady's estimate, not on a bill", gives the estimate and the budget, and says that nothing was sent. Your message stays in the box.
- It starts again with the next calendar month, or when you raise the amount, or when you set it to 0, which turns the budget off.
Not in the month's estimate, though your provider may charge for them: a request you stopped, a request that failed on its way, the small request Test sends in Settings, and a request to a model without prices (Settings says how many of those there were). A server on your own PC without prices is not counted at all.
Turning a limit on
The first time you set an amount for either limit, and each time you turn one on again after 0, a dialog says what you are turning on: the limit is counted from RigReady's estimates at the prices in Settings, it is not your provider's bill, and the limit that holds is the one at your provider. Turn on the limit can be pressed once you ticked that you understand this; Cancel changes nothing. Settings then shows the day you acknowledged it.
Cache and the end of a question
With Anthropic, the part of a request that did not change since the one before (the instructions, the tools, the conversation so far) is read from the provider's cache at the cache read price when the requests follow each other within five minutes. A question always ends once the assistant has used tools 16 times for it.
When a request fails
If the provider refuses a request or cannot be reached, the panel says why, your message is back in the message box, and the request stays ready: Try again sends it again. If you store another key or change who answers in Settings first, the request is made again for the new settings and shown before anything goes.
Every request a provider refused leaves a line in RigReady's log: the host, the model, the HTTP status, the provider's own error and its request id. The key, your messages, what the tools answered and the replies are never written to the log.
It answers from this manual
The assistant has this manual at hand. For a how-to question, and before it proposes a change, it searches the manual (manual_search), reads the page or the section that fits (manual_read), answers from it and links the page, for example Scripts as https://rigready.io/docs/setups/scripts/. When the manual does not cover something, it says so.
Build a ready check
Ask in plain words: "Set up a DCS session with SimAppPro, TrackIR and SRS". The assistant:
- Finds where the programs are installed with the tool
programs_find. It looks at the folder names under Program Files, Program Files (x86), the Programs folder of LocalAppData and your Steam libraries, and at the uninstall entries in the registry, and only reads. It knows the folders the usual names hide (SRS is DCS-SimpleRadio-Standalone, TrackIR may be under NaturalPoint, G HUB is LGHUB). Paths come back with path variables, as a setup keeps them:{PROGRAM_FILES_X86}/TrackIR5/TrackIR5.exe.
- Proposes the setup with
setup_propose: a new setup, or checks and launch steps added to one you have. The proposal appears in the conversation with every check, its fix and every step. Show the setup as it would be saved shows all of it. Save writes it the way the setup editor does; Discard throws it away. Nothing is saved before you press Save, and a setup that changed after the assistant read it is not overwritten.
After saving, Open in the setup editor takes you to the setup, where you can change anything, for example what Launch starts.
To answer what a setup checks or starts, the assistant reads that one setup (setup_read), and before it writes a check or a step it looks up the check and fix types it needs (setup_types). It can also explain why the rig is not ready (rig_check, check_explain), run Make ready or Launch (asked in RigReady first, unless you chose otherwise in Settings; see Safety), and reach every other action of RigReady through rigready_call.
What is on the Safety page, and what is not
A setup or a script saved from the assistant's draft is a file in RigReady's own data folder. Like a setup you save in the setup editor, it is not listed on the Safety page: that page lists changes to files outside RigReady's folder, each backed up first, and setups you deleted. To take a saved proposal back, edit or delete the setup in the setup editor.
A change the assistant makes through a tool (for example writing bindings, or a fix that Make ready runs) is asked in RigReady's window first, unless Settings says not to ask, and a file it changes outside RigReady's folder is backed up and listed on the Safety page under "Built-in assistant".
Scripts
When no check type does what is needed, ask for a script: "Write a script to start SRS minimized as the fix of the SRS radio check". The assistant drafts a PowerShell (.ps1) or batch (.cmd) file with script_draft, with an explanation in plain words. The panel shows the whole file exactly as it would be saved, the explanation, where it goes in the setup, and a warning for anything worth a second look (it reaches the internet, deletes files, closes programs, changes the registry). A script that asks for administrator rights is refused before you see it.
Save writes it into RigReady's data folder, in scripts\assistant, with a header that says an assistant wrote it, and records its SHA-256 as your approval. Then it is added to the setup with Ask before running on, which cannot be switched off for a script an assistant wrote. From then on:
- Before it runs, RigReady asks Run this program? with its path and arguments, every time.
- If the file changes after you approved it (edited by hand, or by another program), RigReady does not run it: the fix or check says it "has changed since you approved it". The panel shows the script as it is now, and Approve it as shown approves exactly that text.
- No assistant can run it. A script an assistant wrote runs only from RigReady's window, after you were asked.
- It runs as you, never as administrator.
- It is never part of a setup you share.
More on how scripts run in Scripts.
What it cannot do
It cannot see your API key or any other secret, change the AI or MCP settings, open file dialogs, use your clipboard, answer RigReady's questions for you, save a proposal or a script, or run a script it wrote. It does nothing while you are not using the panel.