Getting Started

Welcome to MultipleChat

MultipleChat runs ChatGPT, Claude, Gemini, and Grok together — coordinating them into structured multi-step pipelines so they build on, verify, debate, or challenge each other's answers.

AI Collaboration

Assign models to slots, pick a mode — Smart, Debate, Humanize, Expert, and more. Each mode is a different orchestration pipeline.

How it works →

Studio Tools

Generate Word documents, Excel spreadsheets, and PowerPoint presentations from natural language — with live preview and iterative refinement.

Explore Studios →

Projects

Upload files or import them from Google Drive, GitHub or Dropbox to give every chat in a project persistent context — no copy-pasting required.

Learn about Projects →

Two Ways to Make Files: Chat or Studios

MultipleChat can produce real Word documents, Excel spreadsheets, and PowerPoint presentations in two places — and they're for different situations:

Make it in chat

Just ask. "Write me a memo about the new policy" or "make me a Q3 budget spreadsheet" — the AI generates the file inline and gives you a download link, without leaving the conversation.

Best when: You want one quick file, you're already deep in a chat, or you want the file to reference the conversation context.

Open a Studio

Studios are dedicated workspaces with templates, style presets, side-by-side previews, refinement panels, and the ability to run two models on the same document at once.

Best when: You want polished output, deliberate styling, multiple revisions, or you're processing a large file you've uploaded.

Both paths produce the same kinds of files (DOCX, XLSX, PPTX, PDF). Pick whichever feels right for the task.

Account Setup#

MultipleChat uses Firebase Authentication by Google for authentication. No password is stored on our servers — Firebase handles the sign-up and login flow.

Creating an Account

1

Visit multiplechat.ai and click Get Started

2

Enter your email and choose a password, or continue with Google or Apple

3

Verify your email address

Check your inbox and spam folder for the verification link

4

Complete onboarding and start chatting

Trying It Without an Account

You can use MultipleChat without signing in to try it out. Anonymous use comes with a small daily allowance of messages and uses a fixed default model setup. To unlock collaboration modes, file uploads, projects, and the Studios, create an account.

Changing Your Password

Because authentication runs through Firebase, password changes go through Firebase's password reset flow — there is no password-change form inside the app.

  1. 1. Go to the login page and click Forgot password
  2. 2. Enter the email address associated with your account
  3. 3. Firebase sends a password reset link to that email
  4. 4. Follow the link to set a new password

If you signed up with Google or Apple, your password is managed by that provider — not Firebase.

Your First Chat#

Single Model Chat

1

Click the model selector next to the send button and pick one model

2

Type your message and press Enter

AI Team (Collaboration) Chat

1

Select AI Team from the model selector — the AI Team configuration modal opens

2

Tap model icons to assign them to slots (slot 1 = first model, slot 2 = second model). Pick a mode from the pill buttons at the bottom of the modal.

3

Click Go Live to save and close the modal

4

Send your message — the selected mode runs and streams the result back token by token

Good first prompts to try

  • • "Explain how central banks control inflation" — try Smart (Chain)
  • • "Should I use microservices or a monolith for my SaaS?" — try Debate
  • • Paste an AI-generated paragraph and ask to rewrite it — try Humanize
  • • "What are the trade-offs of nuclear vs. solar power?" — try Expert
  • • "Make me a pitch deck for a coffee subscription startup" — the AI calls a tool and gives you a real .pptx file

How AI Collaboration Works#

When you send a message in AI Team mode, the selected mode routes your prompt through a specific sequence of model calls. The final model streams its answer back to you token by token, with a live progress indicator while intermediate steps run.

The General Flow

1

Configure your team — assign models to slots and pick a mode in the AI Team modal

2

Slot 1 model processes your prompt and produces an initial response

3

Slot 2 model receives that output and performs the mode's specific task on it

4

The final synthesized answer streams live to your screen

Every mode uses at least two models — slot 1 writes a draft, slot 2 does the mode's specific job on top of it (rewrite, critique, counter, and so on). Conversation and Co-operative can also use the models in slots 3 and 4 as extra reviewers.

All Collaboration Modes#

Each mode is a distinct pipeline. Choose based on what you need from the result.

Smart (Chain)

Slot 1 generates a complete draft. Slot 2 acts as a domain expert — it takes the draft as a foundation, adds depth, reasoning, analogies, and examples, and challenges any facts Slot 1 got wrong. Sources from both models are combined in the final output.

Best for

Comprehensive explanations, research questions, anything requiring depth and accuracy

Flow

Slot 1 drafts → Slot 2 deepens and critiques → streamed final answer

Humanize

The model rewrites the content you provide — it does not answer it. It removes AI clichés ("delve into", "it's important to note", "leveraging"), varies sentence rhythm and length, and preserves every fact and meaning exactly. Paste AI-generated text as your prompt and get back a natural-sounding rewrite.

Best for

Making AI-written content sound human — blogs, reports, emails, marketing copy

Note

Paste the text you want humanized as your prompt — don't ask a question

Debate (Competitive)

Slot 1's answer becomes the opening position. Slot 2 challenges it, Slot 1 defends it, and Slot 2 then writes the final answer from the strongest arguments on both sides. The debate always runs these three rounds — you can read each one under AI Collab Steps.

Best for

Decisions, trade-off analysis, "should I do X or Y" questions

Flow

Slot 1 position → Slot 2 challenge → Slot 1 defense → Slot 2 final answer

Ensemble

Both models answer your question independently — the second model doesn't see the first one's answer. The two answers are then merged, combining the best parts of each into a single, more complete answer.

Best for

Getting broad coverage without one model influencing the other

Flow

Slot 1 and Slot 2 answer independently → answers merged into one

Co-op (Cooperative)

Models build on each other iteratively. Slot 1 writes a complete draft. Slot 2 — plus slots 3 and 4 if you've assigned them — reviews it and suggests improvements. Slot 1 then produces a final version that incorporates those suggestions.

Best for

Writing tasks, code, structured documents where iterative quality improvement matters

Flow

Slot 1 draft → Slot 2 review & suggestions → Slot 1 final version

Simulation

Models collaboratively play out a scenario over multiple rounds. Slot 1 sets up the scenario, defines key actors and variables, and runs the opening round. Slot 2 plays out the reactions, Slot 1 resolves the scenario, and Slot 2 then writes a final analysis of what happened. You can read every round under AI Collab Steps.

Best for

Scenario planning, hypotheticals, roleplay, strategy exercises, creative writing

Flow

Slot 1 setup + Round 1 → Slot 2 Round 2 → alternating rounds → conclusion

Expert

Slot 1 produces an initial analysis. Slot 2 then adds expert-level depth — the why behind the what, technical depth, strategic implications, trade-offs from multiple professional perspectives. The result reads like a domain specialist's take rather than a general explanation.

Best for

Deep technical or strategic questions where surface-level answers aren't enough

Flow

Slot 1 initial analysis → Slot 2 expert depth → final answer

Conversation

Slot 1 writes a draft. The other models on your team (slots 2–4) each reply with a short, substantive pushback, shown as chat bubbles while it runs. Slot 1 then writes the final answer, taking that feedback into account.

Best for

Stress-testing an answer with pushback from several models before you rely on it

Flow

Slot 1 draft → slots 2–4 push back in parallel → Slot 1 final answer

Quick guide: which mode to use

  • • Writing and content: Humanize or Co-op
  • • Deep research and accuracy: Expert or Smart (Chain)
  • • Decisions and trade-offs: Debate (Competitive)
  • • Broadest coverage from multiple perspectives: Ensemble
  • • Domain depth and technical analysis: Expert
  • • Scenarios and hypotheticals: Simulation

Compare Mode#

Compare Mode is different from any of the collaboration modes. Instead of combining models into one answer, it shows you each model's response to your prompt side by side, in its own column, streaming live in parallel. You see exactly where they agree, where they differ, and which model handled the question best.

Turning It On

Click the Compare button in the top toolbar. The screen splits into columns — one per model — and your input bar moves to the bottom. Type a prompt and every selected model answers at once. On phones you see one model at a time with a switcher. Compare Mode is available to every signed-in user; on the Free plan some columns are locked. Each column uses your message allowance.

What You Can Do

Pick which models to compare — toggle ChatGPT, Claude, Gemini, and Grok on or off in the model bar above the columns
Continue with one model — each column has a Reply button so you can keep the conversation going with just that model
Stop everything — abort all running streams at once if you spot what you needed early
Mobile fallback — on phones, the comparison shows one model at a time with a switcher

Compare vs Ensemble

Both use several models at once. Ensemble merges the outputs into a single combined answer — you don't see each model individually. Compare Mode keeps each response separate so you can read, judge, and continue with whichever you prefer.

What the AI Can Do in Chat#

You don't always need to open a Studio. The AI in regular chat can produce real files and analyze your uploads directly — just ask in plain language and the model will hand you back a downloadable file or a structured analysis.

What You Can Ask For

Create a Word document

"Write me a one-page memo to my team about the new policy" — get a real .docx file back, with optional PDF.

Create a spreadsheet

"Make me a Q3 budget tracker with categories and totals" — returns a real .xlsx with formulas, formatting, and charts.

Create a presentation

"Build me a 10-slide pitch deck about a coffee subscription startup" — returns a styled .pptx.

Generate an image

"Draw a flat illustration of a fox reading a book" — image appears inline in the chat.

Modify your document

Upload a DOCX and say "make this more formal" or "add an executive summary" — get back the rewritten file.

Analyze your file

Upload a long PDF or large spreadsheet and ask questions — the AI pulls the right parts to answer accurately.

Convert a PDF to Word

Upload a PDF and ask for it as a Word file — you get a word-for-word .docx conversion, not a summary.

Create or edit code files

Ask for a script or config file, or upload source code and ask for changes — you get back a file with the same extension.

In-chat vs Studios — when to use which

In-chat is fastest when you want one quick file inside a conversation. The Studios are better when you want to iteratively refine, pick templates and styles deliberately, run two models in parallel for the same document, or process larger files with more options. Both produce the same kinds of files.

Code Sandboxes (ChatGPT & Claude)#

When you work with files, you can choose who processes them. Besides MultipleChat's own file tools, you can send files to a secure code sandbox run by OpenAI (ChatGPT) or Anthropic (Claude). In a sandbox, the AI writes and runs real Python code on your files: it calculates exactly, checks data, draws charts, and hands you finished files to download.

Choosing How Files Are Processed

The first time you attach a file, MultipleChat asks how you want to work with files. You can change this at any time in Account → File preferences, or re-run a single answer with a different processor using the Sandbox button next to Retry.

Process with MultipleChat (default)

Our own analysis, OCR, conversion and file tools, with any chat model or AI Team. Your files are not sent to a provider sandbox.

Send to ChatGPT sandbox

ChatGPT runs Python in OpenAI's Code Interpreter. Your files are uploaded to OpenAI for that turn. The chat switches to a single ChatGPT model.

Process in Claude sandbox

Claude runs Python and shell commands in Anthropic's code execution sandbox. With a Claude chat, Claude does everything. With a ChatGPT chat, ChatGPT keeps the conversation and hands the coding work to Claude; then both OpenAI and Anthropic are involved and both providers' usage applies.

Sandboxes require a paid plan. Your saved default or chat selection applies across topics in supported single-model chats; the AI decides when code is needed. If you have not chosen a processor, we ask before sending or on the first upload. Choosing Not now uses MultipleChat. Message wording does not switch your processor. Smart, Compare and unsupported models use MultipleChat. Select MultipleChat for web search and our document and image tools.

What Sandboxes Can Do

Exact calculations and data analysis

Totals, statistics, cleaning, filtering, joining and reshaping tables, simulations — computed by code instead of estimated by the model. Available libraries include pandas, numpy, scipy, scikit-learn and statsmodels (Claude) and a Python data stack (ChatGPT).

Charts that the AI checks

Bar, line, scatter and other charts (matplotlib, seaborn). After creating a chart, the AI looks at the actual image, fixes layout problems such as clipped labels, and checks again. Ask for a specific style (colours, fonts, minimal look) for a more polished result.

Real files to download

CSV, TSV, TXT, Markdown, JSON, PDF, Excel (XLSX), Word (DOCX), PowerPoint (PPTX) and images (PNG, JPEG, WebP, GIF). Files are checked before download; macros, executables and other formats are not offered as downloads (scripts and HTML are shown in the answer instead). Each final file appears as a download card in the chat. If a revised version fails to save, the last working version stays available and is labelled as the previous version.

Working on your files

Your original chat and project files are placed in the sandbox (for example CSV, Excel, JSON, XML, text, PDF and images, depending on the provider). The AI can read, combine, convert and transform them, and report only the checks it actually ran.

Diagrams and code-drawn graphics

Infographics, flow diagrams, simple vector graphics and image edits made with code (for example with Pillow). For artistic illustrations or photos, use Image Studio instead — sandboxes do not generate images with an image model.

Run your code

Paste or upload a Python script and ask the AI to run, test or fix it. The AI can rewrite and rerun failing code until it works.

Limits

  • No internet inside the sandbox. Code cannot download data or install extra packages; only pre-installed libraries are available. (Claude can still search the web for your answer outside the sandbox.)
  • Files per message: each message can include several input files, within per-file and total size limits. Generated downloads have a size limit too. Files that are too large are not placed in the sandbox, and the AI is told so.
  • Time: each sandbox answer has a time limit. When ChatGPT hands work to Claude, only a small number of separate tasks run per answer, each within the remaining time.
  • Computing resources: sandboxes have limited processing power, memory and disk space. Very large datasets may exceed these limits.
  • Continue in the same workspace: supported sandbox chats reuse their provider workspace across messages and page reloads. OpenAI can retain Python state; Claude can retain working files, while separate Python processes may not share variables. Provider sessions expire after a period of inactivity. We restore available saved files when starting fresh and tell you when temporary state is lost.
  • Interrupted work is not repeated automatically. If a sandbox answer stops, you can retry with the same processor or choose another one; retrying or switching uses additional AI usage.

Privacy and Data Handling

  • When you choose a sandbox, the files and instructions needed for the task are sent to the provider you chose: OpenAI for the ChatGPT sandbox, Anthropic for the Claude sandbox, and both when ChatGPT hands work to Claude.
  • Input files uploaded for signed-in users may be reused for a short time and are configured to expire at the provider soon after. We recheck access and file contents before reuse; these cached uploads are not deleted after each answer. Providers may keep sandbox data for a limited time according to their own policies.
  • Generated files are stored with your chat in your MultipleChat account.
  • Do not upload data that you are not allowed to share with these providers. If you do not want files to leave MultipleChat's own processing, keep Process with MultipleChat selected.

Tips

  • Say exactly what you want back, for example: "Create totals.xlsx and a labelled bar chart chart.png".
  • Ask the AI to "report only the checks you actually ran" when accuracy matters.
  • Open spreadsheets in Excel or another calculating viewer; some totals may be stored as formulas that simple previewers do not calculate.

Disclaimer — no liability for sandbox results

Code sandboxes are provided by third parties (OpenAI and Anthropic) and are offered "as is" and "as available". Code, calculations, charts, files and analyses produced in a sandbox are AI-generated and may be incomplete or wrong, even when the AI reports that it checked them. You are responsible for reviewing every result and every downloaded file before you rely on, share or open it, and for having the right to send your files to the provider you choose. To the fullest extent permitted by applicable law, MultipleChat is not liable for any loss or damage arising from sandbox outputs, from code run in a sandbox, from files generated or downloaded through a sandbox, or from the availability, behaviour, data handling or retention of third-party providers. Generated content must not be used without independent expert review for legal, medical, financial, engineering or other high-risk purposes. See our Terms of Service, in particular sections 4.2, 14.4, 15 and 25, which apply to sandbox use.

Document Studio#

Document Studio generates and processes Word documents (.docx) from natural language. Generate from scratch using a template, upload an existing file to transform it, refine the result iteratively, and preview the output as a live HTML preview before downloading.

Why a Studio instead of just chat? Document Studio gives you templates with built-in section structures, side-by-side competitive drafts from two models, a quality-review summary, a live preview before download, and a refine panel to iterate on the result. Chat is faster for one-shot files; Studio is better when the document matters.

Document Templates

Each template defines a structural blueprint — the sections the AI will populate:

Business Report — executive summary, findings, analysis, recommendations
Proposal — problem statement, solution, methodology, timeline, budget
Whitepaper — abstract, background, methodology, findings, references
Case Study — challenge, approach, implementation, results, lessons learned
Statement of Work — scope, deliverables, milestones, acceptance criteria
Meeting Minutes — attendees, agenda, discussion, decisions, action items
Cover Letter — opening, qualifications, value proposition, closing
Resume / CV — summary, experience, skills, education, certifications
Internal Memo — purpose, background, discussion, action required
Creative / Essay — introduction, body sections, conclusion
Custom / Freeform — AI decides the best structure from your topic and instructions

Collaboration Modes

Document Studio has its own generation collaboration modes (separate from the main chat modes):

Single Model

One AI model generates the complete document.

Cooperative Writing

Model A drafts → Model B reviews and suggests improvements → Model A produces the final version. After generation, you'll see a quality summary showing strengths, weaknesses, and how many improvements were applied.

Competitive Drafting

Two models each produce a full, independent draft. You see both side by side and can pick or merge the best parts.

Processing an Existing Document

Upload a DOCX, PDF, or TXT file and apply one of these actions:

Improve — fix grammar, enhance clarity and professionalism
Summarize — condense into a shorter version keeping key points
Expand — add detail, examples, supporting evidence
Restructure — reorganize for better flow and logical progression
Tone Adjust — shift to formal, casual, technical, or persuasive
Extract Actions — pull all action items, decisions, and takeaways
Merge — combine multiple source documents into one

Export Formats

Every generated document can be downloaded as DOCX, PDF, Markdown (.md), or HTML. A live HTML preview is shown before you download anything.

Iterative Refinement

After generating, use the Refine panel to describe what you want changed in plain language. The studio applies your instructions and updates the live HTML preview. Repeat as many times as needed before downloading.

Excel Studio#

Excel Studio is an AI-powered spreadsheet environment. It generates workbooks from scratch, lets you refine them with plain-language instructions, profiles files you upload, runs free instant tools that need no AI, and includes a Formula Assistant that writes and explains any Excel formula on demand.

Why a Studio instead of just chat? Excel Studio profiles your data on upload (rows, missing values, per-column stats), runs free instant tools (cleaning, formatting, formulas, reshaping) without using AI credits, and includes a dedicated Formula Assistant. Chat can make a spreadsheet; Studio is for working with one.

1. Generate a Spreadsheet from Scratch

Describe what you need in plain language. Choose a template, a style preset, and how many rows you want. The AI builds a realistic workbook with proper formulas, formatted columns, summary rows, and charts where appropriate — all in one step.

Available templates:

Budget Tracker — income/expense categories, subtotals, variance
Profit & Loss Statement — revenue, COGS, expenses, net income, % of revenue
Cash Flow Forecast — opening/closing balance, inflows, outflows, runway
Loan Amortization — PMT schedule with principal, interest, remaining balance
Invoice — line items, quantity, unit price, tax, total
Sales Pipeline / CRM — deals, stages, probability, weighted forecast
Employee Roster — departments, salary, tenure, next review date
Inventory Tracker — SKUs, reorder point, unit cost, total value
Project Tracker — tasks, owner, dates, status, dependencies
OKR / KPI Scorecard — objectives, key results, targets, traffic-light status
Comparison Matrix — options vs weighted criteria, scoring, ranking
Expense Report — dated expenses, categories, vendor, totals by category
Custom / Freeform — AI decides the best structure for your description

2. Work With an Existing Spreadsheet

Upload an existing .xlsx or .csv file. The studio automatically profiles your data — row count, column count, duplicate rows, missing values, and per-column statistics (min, max, mean for numeric columns). To change the file, use the free instant tools in section 5 below: they clean, format, add formulas and reshape your data without using AI credits.

Process File: On Pro or Smart, describe supported cleaning, built-in formula, formatting or reshape operations. Choose the worksheet to change, then review the change summary and download a new workbook. Upload XLSX, CSV or TSV files within the size limits shown in the studio. Translation, chart creation and arbitrary formulas are not supported in this workflow. Operations that could break workbook references are rejected, and failed processing jobs refund their reserved tokens.

3. Iterative Refinement

After generating a spreadsheet, use the Refine panel to describe changes in plain language — "Add a column that calculates 30-day average", "Change the header colours to dark blue", "Add a pie chart to the Summary sheet". The AI modifies only the parts you asked about and rebuilds the file. Repeat as many times as needed.

4. Formula Assistant

The Formula Assistant is a dedicated tab that writes Excel formulas on demand. Describe what you want in plain English, optionally paste your column headers as context, and get back:

  • • The exact formula ready to paste into a cell
  • • A step-by-step explanation of how it works
  • • Edge-case warnings (e.g. what happens when there's no match)
  • • Alternative formulas for the same result

Example requests:

  • • "Find the last purchase date for each customer in column A, looking up column B in the orders sheet"
  • • "Sum column D only where column C says 'Approved' and column E is greater than 1000"
  • • "Calculate the number of working days between two dates excluding weekends"

5. Free Tools (no AI, instant)

These run instantly — no AI call needed. Upload a file on the Tools tab and click any card:

Trim whitespace
Remove duplicate rows
Standardize text case
Fix date formats
Split column by delimiter
Merge columns
Remove empty rows/cols
Forward/back-fill blanks
Add summary row
Running total column
% of total column
Period-over-period growth
Rank column
Auto-format as table
Color scale formatting
Data bar formatting
Freeze header row
Auto-fix number formats
Auto-fit column widths
Pivot/summarize to sheet
Transpose rows/columns
Split by category column

Export Formats

Download as Excel (.xlsx), CSV, or TSV. A live HTML preview of the spreadsheet is shown before downloading.

Presentation Studio#

Presentation Studio generates and processes .pptx files using a full design system. You control the style template, colour palette (preset or custom hex), font pairing, slide count, and can provide source content or instructions. Jobs run in the background with a live progress bar, and the finished file auto-downloads when ready.

Why a Studio instead of just chat? Presentation Studio gives you style templates, colour presets (or custom hex), font pairings, a precise slide-count control, and the ability to upload an existing deck and transform it (summarize, expand, redesign). Chat picks defaults for you; Studio puts you in control of how the deck looks.

1. Generate a Presentation

Enter a topic, pick your settings, and click Generate. You can also paste source content — notes, an article, a brief — and the AI will structure slides from it rather than generating from scratch.

Style Templates

Business — professional, clean layout
Pitch Deck — investor-focused structure
Educational — clear sections, training-ready
Creative — bold visuals and layouts
Report — data-driven, metrics-focused
Minimal — content-first, low decoration

Colour Presets

Choose from built-in colour schemes — Corporate, Modern Dark, Nature, Warm, Ocean, Sunset, Monochrome, Pastel — or override the primary, secondary, and accent colours with your own hex values using the colour pickers. A Reset button reverts custom colours to the active preset.

Font Pairings

Classic — Georgia / Calibri
Modern — Arial Black / Arial
Elegant — Garamond / Cambria
Clean — Helvetica / Helvetica
Bold — Impact / Verdana
Friendly — Trebuchet MS / Tahoma

Slide Layouts

The AI mixes layout types to avoid repetition. Every presentation always opens with a Title slide and ends with a Closing slide. Speaker notes are generated for every slide.

Title
Content
Two Column
Section
Stats
Quote
Closing

2. Process an Existing Presentation

Upload a .pptx file or pick one from your Projects (no need to re-upload anything you've already stored there). Apply an action with optional free-text instructions:

Summarize — condense into fewer, more impactful slides
Expand — add detail and supporting points to each slide
Redesign — restructure narrative flow, swap layouts for better impact
Improve — fix grammar, sharpen bullet points, tighten writing
Add Speaker Notes — generate comprehensive notes for every slide
Reorder — reorder slides for better narrative flow and logical progression
Custom — describe anything in plain language

Processing builds a new deck from the text of your slides — the original design, images and animations aren't carried over.

3. Download & Auto-Download

When a job completes, the .pptx file auto-downloads immediately. A result card also appears showing the presentation title, slide count, model used, and a manual download link in case the auto-download was blocked.

Jobs run in the background

Generation and processing happen in the background — you can keep working while a job runs and a real-time progress bar updates as it moves through stages. The page shows the result as soon as it's ready.

Image Generator#

The Image Generator is a dedicated page (/image-generator) for creating images from text prompts. It gives you a choice of frontier image models and several aspect ratios.

Image Generation Models

Pick one or more models in the model selector. Some are strongest at following detailed instructions, some at photorealism, some at stylized and cinematic art, and some at legible text inside images such as posters and logos. Try the same prompt on several models and keep the result you like best. Available models can change and may depend on your plan.

Generation Options

Aspect Ratios

1:1
16:9
9:16
4:3
3:4
21:9

Writing Good Prompts

  • • Be specific about subject, setting, and mood
  • • Name a visual style: photorealistic, oil painting, 3D render, flat illustration
  • • Add technical detail: golden hour lighting, wide angle, shallow depth of field

Image Studio#

Image Studio is a separate page (/image-studio) for working with images you already have. Upload a photo or graphic and edit it with plain-language instructions, or remove its background with one click. Whereas the Image Generator creates images from text, Image Studio transforms images you provide.

What You Can Do

Upload an image

Drop a PNG, JPG, or WEBP file into the studio. The image becomes the base for everything that follows.

AI edit

Describe a change — "make this look like a watercolour", "change the background to a beach", "add sunglasses" — and the AI returns a modified version.

Remove background

Click the Remove Background button — the subject is automatically isolated and you get back a transparent PNG ready to drop into a design.

Download

Save the result as PNG or JPG. The original is preserved alongside the edits in the session.

Image Studio is for editing existing images. To create images from scratch, use the Image Generator.

Prompt Optimizer#

The Prompt Optimizer rewrites your message before you send it, making it clearer, more specific, and more actionable for AI models — without changing your intent or inventing new facts. It preserves URLs, numbers, filenames, and code blocks exactly as written.

How to Use It

1

Type your prompt in the message input. The Optimize button activates once you've typed something.

2

Click the Optimize button. A modal opens showing the rewritten prompt in an editable textarea.

3

Below the rewritten prompt, enrichment chips appear — clickable tags like Audience, Tone, Length, Structure, Images, Sources. Clicking a chip appends a placeholder line to the prompt that you can fill in.

4

Edit the optimized text directly if needed, then click Use to replace your original message — or Use & Send to replace and send immediately.

What the optimizer does

  • • Rewrites vague prompts into clear, imperative instructions
  • • Expands minimal prompts just enough to make them actionable — without inventing specifics
  • • Preserves your language — the output stays in the same language as your input
  • • Does not change facts, URLs, numbers, or code

Auto-Verification#

After a response is delivered, MultipleChat can automatically run it through a factual accuracy check by a separate frontier model. The check analyzes whether the answer contains factual errors, outdated information, or missing context — and shows a badge on the response.

How It Works

The checking model receives your original question and the AI's response. In one step it decides whether verification is appropriate for this type of content and, if so, what is factually correct and what issues exist. The result appears as a Verified badge if no issues are found, or an Issues found badge with a brief description if problems are detected.

Smart Skip Logic

Verification is automatically skipped — saving credits — for content types where factual checking doesn't add value:

Simple math and calculations
Code generation and debugging
Creative writing tasks
Opinions and subjective questions
How-to instructions
Definitions and grammar help

Auto-Verification is off by default. Turn on Auto-verify answers in the tools menu next to the message box (paid plans). Once it's on, every answer is checked right after it's delivered, and each check uses your message allowance.

AI Disagreements#

In collaboration modes, the model that writes the final answer is also instructed to review the draft and list the issues it found. They are stripped from the visible response and shown in a separate AI Disagreements panel under the final answer.

What Gets Flagged

The reviewing model looks for issues in five categories:

Factual Errors

Incorrect dates, numbers, names, or claims

Missing Context

Important details that were left out

Outdated Information

Things that may have changed or been superseded

Oversimplifications

Nuance or complexity that was lost

Better Alternatives

A better approach, tool or option the answer didn't mention

Each disagreement is categorised and displayed with a colour-coded icon — red for factual errors, amber for missing items, orange for outdated info, blue for context issues. Click AI Disagreements under any collaboration response to open the panel.

The disagreements block is always stripped from the final answer text — it never appears inline in the response. It only surfaces when you explicitly click to view it.

Sources#

When models use the web or cite references, those sources are collected and surfaced as a Sources button under the response. Depending on the mode, they come from one or both models, with duplicates removed.

Clicking Sources opens a modal showing each reference with its title, domain, and a clickable link. Some links are cited by the model itself, so open and check any source you plan to rely on.

Like disagreements, sources never appear inline in the readable response. The Sources button only appears when sources were actually found.

AI Collab Steps#

In collaboration modes, the final answer you see is the synthesized output of multiple models working together. To see how it was built, click the AI Collab Steps icon under a collaboration answer.

This opens a panel with the intermediate steps of the mode you used — for example the first draft, the reviewers' feedback, the debate rounds or the simulation rounds. It is useful when you want to understand exactly how the final answer was built.

For simple queries where the collaboration step was skipped to save time, the AI Collab Steps panel will show a note explaining that a quick single-model response was used instead.

Team Reasoning#

Click the Reason button under any answer to get a critical second look at it. The model picks the most useful angle — the root cause behind a symptom, a stress test against edge cases, a way to reconcile two valid views, the fit with your actual situation, or a known pattern from the field — and points out what the answer missed, hedged or assumed.

Team Reasoning is available on paid plans and uses your message allowance each time you run it. The result isn't saved with the chat — click the button again to re-run it.

Projects#

A project is a named workspace that bundles files and standing instructions together. The instructions apply to every chat in the project, and the AI can look through the project's files whenever a question needs them — no copy-pasting required.

Creating a Project

1

Click New Project from the Projects page

2

Name the project and add optional instructions

Instructions are a standing system prompt — they shape how the AI behaves in every conversation inside this project

3

Upload files or import them from Google Drive, GitHub or Dropbox

Supported: PDF, DOCX, TXT, MD, CSV, JSON, and most plain-text code formats. Archives (.zip, .tar, .gz) are not supported.

4

Start a new chat inside the project — the AI follows your instructions and can read your project files

Google Drive, GitHub & Dropbox

Connect your Google, GitHub or Dropbox account to import files straight into a project — pick files or folders from Drive, files from Dropbox, or files from any GitHub repository your account can access. Use Re-sync any time a file changes — syncing is triggered manually so you control when files are updated.

How context retrieval works

Your project instructions are included with every message. When a question needs your files, the AI searches the indexed files and reads the relevant parts — not entire files. Search matches words, so naming the file or using the terms from your documents gives the best results; you can also simply ask the AI to check your project files. If files have not been indexed yet, a banner appears with a one-click Index Now button.

File Uploads#

You can upload files directly in any chat — outside of a project — to give the AI context for that conversation. Click the paperclip icon next to the message input, or drag and drop a file into the chat window.

Supported Formats

Documents

PDF, DOCX, XLSX, PPTX, TXT, MD, CSV, JSON, XML

Images

PNG, JPG, GIF, WEBP — analysed visually by multimodal models

Code

JS, Python, HTML, CSS, and most plain-text code formats

How Files Are Read

MultipleChat picks the best path automatically based on the file:

  • • Small text and code files — injected directly into the conversation context so the AI can quote and reason over the whole content
  • • Spreadsheets, PDFs, large documents, and presentations — automatically routed to a Studio-style structured analysis so the AI can extract tables, sections, and statistics rather than treating them as walls of text
  • • Scanned PDFs — automatically OCR'd at upload time so handwritten or image-based PDFs are still readable

Per-plan upload limits apply to file size, total upload size, and how many files you can attach in a single message. Files uploaded into individual chats also expire after a period — long-term storage belongs in Projects.

Uploaded file contents are treated as reference data, not as instructions to the AI. Even so, only upload files you trust; see prompt injection. Very large files and long chats are limited by the model's context window.

Account & Billing#

Your account page (/account) shows your current plan, today's usage, and billing history. All subscription management happens here without contacting support.

Viewing Your Usage

The account page shows progress bars for how many collaboration (AI Team) messages and single-model messages you have used today against your daily limit. Limits reset every day.

Changing Your Plan

Go to Account → Subscription. Upgrade and downgrade options are shown based on your current plan. You are redirected to Stripe for payment and then returned to the app. Changes take effect immediately.

Billing Portal

Click Manage Billing on the account page to open the Stripe Billing Portal. From there you can:

  • • Update your payment method
  • • Download past invoices
  • • Switch between monthly and annual billing
  • • Cancel your subscription

Cancelling

Cancellation is handled through the Stripe Billing Portal. Access continues until the end of the current billing period — nothing is cut off immediately on cancellation.

Subscribed in the mobile app?

If you subscribed in the MultipleChat iOS or Android app, the App Store or Google Play bills you. Manage or cancel that subscription in your phone's subscription settings — not in the Stripe portal.

Chat History#

All your conversations are saved automatically and accessible from the chat history icon in the sidebar.

Managing Chats

  • • Rename — click a chat's title to edit it
  • • Pin — pin important chats to keep them at the top
  • • Delete — remove individual chats
  • • Delete all — clear your entire history at once

Exporting Your History

  • • Use the Export button in the history view
  • • Choose format: JSON, TXT
  • • Export your full history or select specific chats
  • • Import Chats brings in MultipleChat or ChatGPT exports (JSON or CSV)

Privacy

Chat history is private to your account. Users who are not logged in do not have their chats saved between sessions. See our Privacy Policy for full details on data handling.

Sharing & Forking Chats#

If you're on a team, you can share any of your chats with your teammates. Teammates who open the link — signed in and on the same team — see the full conversation read-only, and can fork it into their own chat history to continue in their own direction.

Sharing a Chat

1

Open any chat from your history

2

Click the Share action — a dialog opens with a shareable URL

3

Copy the URL and send it to a teammate — only signed-in members of your team can open it

4

View counts and fork counts are shown in the dialog so you can see how the chat is being used

Forking a Shared Chat

When a teammate opens a shared chat link they can:

Read it — full conversation, including any AI Team configuration that was used
Fork it — clones the chat into their own history under "Fork of [original title]" so they can keep going from where you left off

Unsharing

Open the same Share dialog and click Unshare. The link stops working immediately — anyone who already forked the chat keeps their copy, but no new viewers can access the original.

If you leave the team, all of your shared links stop working automatically.

Files attached to the original chat are not copied into forks — only the conversation text. If a forked chat references files, the new owner will need to upload their own copies.

Teams#

A team account puts several people on one subscription. The owner manages seats and billing on the Team page (/team) and invites members by email.

  • • Seats — the owner adds or removes seats; extra seats are billed right away, prorated
  • • Invites — invite links expire after a few days; joining a team cancels the member's own paid subscription
  • • Team projects — projects created as team projects are shared, and every member can read and edit their files
  • • Chat sharing — members can share chats with each other (see Sharing & Forking Chats)

Usage limits are per member, not pooled — each member gets the team plan's full allowance. Chats and chat uploads stay private unless you share them.

Key AI Terms#

Plain-language definitions of the AI terms you will see in MultipleChat, and what each one means for how you use it. For more than 70 further terms, see our AI Glossary.

AI model (LLM)

A large language model (LLM) is an AI system trained on large amounts of text and code to understand and write language. MultipleChat gives you models from several providers, such as OpenAI (ChatGPT) and Anthropic (Claude). Each model has different strengths, so their answers can differ — see AI Disagreements. More in the AI Glossary: LLM.

Token

Models read and write text in small pieces called tokens; a token is roughly a short word or part of a word. Usage limits and Studio allowances are measured in tokens, so long messages, large files and long answers use more of your allowance. See Account & Billing and the definition in our Terms of Service. More in the AI Glossary: token.

Context window

The maximum amount a model can consider at once: instructions, the conversation so far, attached files and its own answer. When a long chat nears MultipleChat's working limit, older turns are automatically condensed into notes so the chat can continue. Recent messages stay word for word, and your full original history stays saved in Chat History. Details from early in a very long chat may therefore be summarized rather than quoted exactly. A single message or file too large to fit returns a size error; it is never silently cut. More in the AI Glossary: context window.

Knowledge cutoff

Each model learns from data up to a certain date and knows nothing after it unless it searches the internet. The date differs by model. For recent events, prices or releases, use Web Search.

Hallucination

When a model states something false with confidence, such as an invented fact, quote, citation or number. Every model can do this. Check important answers against their Sources, compare models, or use Auto-Verification. See also section 15.2 of our Terms of Service. More in the AI Glossary: hallucination.

Reasoning ("thinking") models

Some models work through a problem step by step before they answer. This usually helps with maths, code and complex analysis, but answers take longer and can use more tokens. More in the AI Glossary: chain-of-thought reasoning.

AI agent and tool use

An AI that takes actions instead of only writing text, for example searching the web, reading your files, or writing and running code in a Code Sandbox. The AI chooses the individual steps itself, so review what it did and the files it produced before you rely on them. More in the AI Glossary: AI agent and tool use.

Prompt injection

Hidden or malicious instructions placed inside content the AI reads, such as a webpage, an uploaded document or a shared file, that try to change what the AI does. MultipleChat tells models to treat uploaded and Project files as reference data, not as instructions, but no safeguard is perfect. Only upload files you trust, be careful with web results, and question any answer that suddenly changes course or asks for personal information, passwords or payments. More in the AI Glossary: prompt injection.

Memory

MultipleChat remembers the conversation inside each chat. It does not automatically carry what you said in one chat into another. To reuse files and context across chats, keep them in a Project.

Your data and model training

MultipleChat does not use your chats or files to train AI models. The provider of the model you choose processes your message to generate the answer, and some providers keep API data for a limited period (typically up to 30 days) for abuse monitoring. Details are in our Privacy Policy.

Troubleshooting#

Models not responding

  • • Check your internet connection
  • • Refresh the page and try again
  • • Try a different browser or disable extensions
  • • Check for known outages on our status page

AI Team mode not working

  • • Make sure you have at least 2 models assigned (slots 1 and 2) in the AI Team modal
  • • Click Go Live to apply your configuration before sending
  • • Check the model selector shows "AI Team" — if it shows a single model name, collaboration is disabled

Compare Mode unavailable

  • • On phones, Compare Mode shows one model at a time — use the switcher to change model
  • • You need to be signed in; on the Free plan some columns are locked — upgrade to use them
  • • Make sure at least two models are toggled on in the model bar before sending

Project files not being used

  • • If you see a "Files not yet indexed" banner, click Index now before sending your message
  • • After adding new files, use Re-sync to update the index
  • • Archives (.zip, .tar, .gz) are not supported — extract the files first

Slow responses

  • • Collaboration modes (especially Debate and Simulation) make several AI calls in a row — they are slower by design
  • • Turn off web search if you don't need real-time information
  • • In Conversation and Co-op, each extra reviewer model adds time — remove reviewers you don't need

Studio job failed or stuck

  • • Studio jobs run in the background — wait for the progress indicator to complete fully before downloading
  • • If a job errors, try again with a simpler or shorter prompt to isolate the issue
  • • Large file uploads (PDFs, DOCX) take longer to process — the status updates in real time

Shared chat link not working

  • • The owner may have unshared the chat or left the team — ask them to re-share
  • • You must be signed in and on the same team as the person who shared it
  • • Check that you copied the full URL, including the token after /shared/
  • • If you're trying to fork and nothing happens, refresh the page once and try again

Still having issues?

Contact us at [email protected]. Include your browser, what you were doing when the issue occurred, and any error messages you saw.