Chat engine
AI chat across Claude, GPT, Gemini and Grok — in one place
One thread, four model families — Claude, GPT, Gemini and Grok — twelve tiers, and modes that argue back. Watch the flow, then run it yourself.
Twelve AI model tiers in one chat thread
Pick the model per message: a quick one for small questions, a careful one for a contract, all in the same conversation.
| Family | Vendor | Tiers |
|---|---|---|
| GPT | OpenAI | GPT-4o Mini, GPT 6 Luna, GPT 6.1 Sol, GPT-6 Astra |
| Claude | Anthropic | Claude Haiku 4.5, Claude Sonnet 5.5, Claude Opus 5.5, Claude Fable 5.1 |
| Gemini | Gemini Flash Lite, Gemini 3.5 Flash, Gemini 3.1 Pro | |
| Grok | xAI | Grok 4.5 |
- One picker, rememberedChoose a family, then a tier. The app reopens on the model you last used.
- Eight tabs that keep runningUp to 8 chats side by side, each with its own mode, model and files, so a long run keeps going.
- Free daily chat on two modelsGPT-4o Mini and Gemini Flash Lite are free for 500 calls a day each in main chat, without web search.
- The same chat on your phoneAn installable web app with the same models, modes, Voice Mode and attachments as desktop.
In detail
Stop a reply, or find the right studio
- Press Cmd or Ctrl plus Backspace to stop a reply mid-answer. A stopped reply is not charged in full.
- Ask chat for an image, a video, music or a website and it points you to the right SynergyAI studio instead of refusing.
The free allowance resets at midnight IST. In Argue, Coach and Consensus those models are priced normally.
Chat with your PDFs, links and the live web
Attach a file, paste a link or switch on web search. Links and search work on all four model families.
- Attach a fileAsk questions about PDFs, Word files, images and code files. The model looks at images directly.
- Paste a linkAn https link is fetched and cleaned, so even a model with no browsing can summarise or quote it.
- Search the live webThe globe turns search on for one message. The sources used appear under the reply as links.
- 10files per message, up to 10 MB each
- 20 MBof attachments per message
- 80,000characters read from a document
- 3links read per message
In detail
File types and limits
- Accepted: .pdf .docx .txt .md .json .csv .js .ts .html .css .py .go .rs .java .c .cpp .png .jpg .jpeg .webp .gif
- PDF and Word files up to 2 MB each.
- Each link adds up to 24,000 characters of page text. Plain http links are not read.
How each model searches
- GPT: a search-preview version of GPT-4o or GPT-4o Mini
- Claude: Anthropic web search, up to 3 uses
- Gemini: Google Search grounding
- Grok: xAI Live Search, up to 3 results
Web search resets after every message, so it never runs unless you turn it on. It is not offered in Argue, Coach or Debate.
Get real Word, Excel and PDF files, plus books and reports
Ask for a spreadsheet and get an .xlsx you can open. Pick a skill, a ready-made long job, and a book, contract or report runs as one job.
- Get PDF, Word, PowerPoint and ExcelThe reply carries the file with Open and Download, on GPT, Claude and Gemini alike.
- Get text and code filescsv, html, md, json, svg, ics, sql and more arrive as real files.
- Revise the file you gotAsk for a change and it reworks that file instead of starting over. Your 50 most recent files sync across devices.
- Write a book10k, 25k, 50k or 80k words, eight genres, four points of view, written chapter by chapter on Claude Sonnet 5.5.
- Draft a document26 document types, from an NDA to a power of attorney, for eight jurisdictions.
- Build a research reportAn overview, deep dive or systematic review, at 500, 1500, 3000 or 5000-plus words.
In detail
How a book run works
- Plan, write each chapter, keep a story bible, check continuity, polish.
- See an estimated range before it starts. It keeps running if you close the tab.
- Stop keeps the work so far. A genuine failure is refunded; a stop you chose is not.
- Regenerate any chapter after a cost confirmation, and export to PDF, Word or ZIP. More in the AI book and story writer.
Document drafts and custom skills
NDAs, employment and service contracts, privacy policies, terms of use, leases, shareholders agreements, MOUs, board resolutions, grant applications, reference letters and power of attorney, among 26 types.
- Jurisdictions: UK, US, EU, Australia, Canada, India, UAE and International
- Lengths: Brief, Standard, Comprehensive and Full document
- A custom skill is an instruction you write once, saved to your account and synced
Formats a text model cannot honestly produce, such as exe, zip, mp3, mp4, png and epub, are refused rather than faked. Skills run on paid model tiers.
Consensus and Debate: several models on one question
When an answer matters, put three models on it and let a judge hand you one verdict to act on.
| Consensus | Debate | |
|---|---|---|
| How it works | Three models answer in parallel; a judge writes one verdict | For, Against and Neutral argue in turn; a judge rules |
| Standard lineup | GPT-4o Mini, Claude Haiku 4.5, Gemini Flash Lite; judged by Claude Haiku 4.5 | GPT 6 Luna, Claude Haiku 4.5, Gemini 3.5 Flash |
| Pro lineup | GPT 6.1 Sol, Claude Opus 5.5, Gemini 3.1 Pro; judged by GPT 6.1 Sol | GPT 6.1 Sol, Claude Sonnet 5.5, Gemini 3.1 Pro |
| Rounds | One, plus an optional Round 2 of rebuttals, priced separately | 1 to 5, and you set how hard speakers push |
| You keep | One verdict you can re-summarise, plus a Markdown transcript | The verdict, a Markdown transcript, and narrated audio as one MP3 |
- Choose each sideGive each model For, Against, Neutral or a stance you write, and edit the motion before it starts.
- Hear it argued aloudNarrated debates give each speaker its own voice, from 30 in American, British and Indian English, and cost more than text debates.
- Join in, or keep goingRecord a turn of up to 60 seconds and the judge decides again, or add rounds after the verdict, priced first.
- Share a verdictPublish a read-only page only after you confirm. Deleting the chat takes it down.
On the phone app, Debate is called Live Debate.
Argue, Coach and Voice Mode: practise against an AI
Test an opinion against a tough opponent, rehearse an interview with a live score, or do either out loud.
- 13Argue opponent personas
- 18Coach role-play scenarios
- 15argument moves Argue opponents draw on
- 8Voice Mode personas
- Argue takes the other sideState a position and a persona argues back, while a Convinced meter tracks how far you have moved it.
- Coach scores you liveIt plays your interviewer or counterpart, scores you out of 100 as you go, and ends with a feedback report.
- Voice Mode talks it outTalk in real time with a live transcript of both sides, in plain chat, Argue or Coach.
- History keeps every sessionArgue and Coach sessions save to Chat History, ready to reopen and discuss.
In detail
The 13 Argue opponents
No persona, Your spouse, Traditional parent, Skeptical boss, The intellectual, Skeptical investor, Curious kid, The lawyer, The journalist, The scientist, Friend as devil's advocate, Frustrated customer, and a Custom persona you write.
- Gentle, Firm or Brutal intensity sets how hard it pushes. The Convinced meter runs from 0 to 100.
- Each persona draws on its own mix of 15 argument moves and avoids replaying the same attacks.
The 18 Coach scenarios
- Tech interview, PM interview, HR and behavioural interview, Consulting case
- MBA admissions interview, UPSC and civil services, Bank PO and government interview, Test prep and oral exam
- VC pitch, Sales pitch and cold call, Salary negotiation
- IELTS and TOEFL speaking, English speaking practice, Foreign language conversation
- Public speaking Q and A, Tough conversation, Dating and first date
- A custom scenario you define yourself
Each scenario has its own scoring dimensions: a VC pitch, for example, is scored on traction, unit economics and moat, among others.
The 8 Voice Mode personas
Default assistant, A close friend, A wise mentor, A witty comedian, A supportive listener, A no-nonsense older sibling, A curious philosopher and A pragmatic coach. Swap persona mid-session.
- Out loud, Argue still shows the Convinced meter and Coach still scores you live.
Agent Mode: hand a job to an agent from the chat
Some requests are jobs, not questions. Agent Mode runs them as missions you cap, approve and collect files from.
- Switch on Agent ModeOne button switches the chat from answering questions to running jobs.
- Set a spending limitSlide a cap for the mission. Nothing is charged upfront; the run is metered against your cap.
- Approve or steerA mission that pauses waits for Approve & continue, Reject, or your guidance.
- Collect the filesBrowse what the run produced. Missions also show up in Workspace, and the other way round.
In detail
Models and limits
- Claude Opus 5.5 runs missions unless you pick another Claude, GPT or Gemini tier.
- Four models just for agents: DeepSeek V4 Pro, Qwen 3.7 Max, Kimi K3 and GLM-5.2.
- Each model has its own minimum spending limit.
- Up to 8 mission tabs, separate from your 8 chat tabs, and up to 8 files of 4 MB each per mission.
- The free tiers switch to a paid tier while Agent Mode is on.
Agent Mode is unavailable in Forget and Burn privacy mode. More on AI agents.
Memory, history and privacy you control
Long threads keep their context, history follows you across devices, and you decide what is kept.
- Long threads rememberOlder messages are recalled when they matter and the rest is summarised. Recalled memory does not change your price.
- History on every deviceEvery Chat, Debate, Argue and Coach session saves itself. Start on your phone, carry on at your laptop.
- Delete means deleteDeleting a chat erases it from the server, not just your screen, and takes down any page shared from it.
- Your apps as contextConnect Gmail, Google Calendar, Drive, Slack, Notion, GitHub, Linear or Dropbox, and chat has a sense of your week.
| Mode | What happens |
|---|---|
| Standard | Sessions are saved; no model of you is built |
| Neural Network | Recommended. An opt-in model of you, learned from chat turns and voice transcripts across the chat, image and video, music and story studios |
| Forget and Burn | Nothing is written to the server; anything transient is deleted after 24 hours. History, sync, saved work and Agent Mode are off |
In detail
See and edit what the Neural Network learned
- Enter the Mind shows your memories as a graph to inspect, pin or forget. A forget can be undone for 30 days.
- Export the graph as JSON, or wipe it in full or in part.
- Sensitive memories and personal identifiers never go into a third-party prompt.
Connected apps in chat
- Chat sees short summaries only: mail subjects, event titles and file names.
- Turning context off or disconnecting an app wipes its summaries.
- App text is treated as data, never as instructions. Connect apps in Workspace.
Pay for what each reply uses
One account covers every studio — top up as you go, or choose a monthly Pro or Studio plan.
- Pay for what a reply usesA short answer costs less than a long one, and each reply shows what it cost once it lands.
- Set a Max per replyChoose a ceiling in the model picker and no reply costs more. A reply that reaches it can carry on with Continue.
- A failed reply is refundedIf a model call fails, the charge comes back automatically and a notice tells you.
- One price per Consensus or Debate runRound 2 and extra Debate rounds are priced separately. A failed call within a run is partly refunded.
- Voice Mode is billed by the minuteEach minute is charged as it begins, with a warning 10 seconds before the next one.
- Argue and Coach work the same wayEach reply costs what it uses, within your Max per reply, and is refunded if it fails.
- Agents run against your capMissions have no upfront charge and are metered against the spending limit you set.
In detail
Safety and limits on every message
- Every message and attached image passes a child-safety check before it reaches any model, and a refusal says why.
- Up to 32,000 characters per message.
- Busy or empty responses are retried on the same model, never silently switched to a different one.
Frequently asked questions
Which AI models can I chat with on SynergyAI?
Four families and twelve tiers. GPT from OpenAI: GPT-4o Mini, GPT 6 Luna, GPT 6.1 Sol and GPT-6 Astra. Claude from Anthropic: Claude Haiku 4.5, Claude Sonnet 5.5, Claude Opus 5.5 and Claude Fable 5.1. Gemini from Google: Gemini Flash Lite, Gemini 3.5 Flash and Gemini 3.1 Pro. Grok from xAI: Grok 4.5.
Is there a free AI chat on SynergyAI?
Yes. GPT-4o Mini and Gemini Flash Lite are free for 500 calls a day each, in main chat without web search. The allowance resets at midnight IST, and every reply shows how many free calls are left. In Argue, Coach and Consensus those models are priced normally.
Can I upload a PDF or a Word document and ask questions about it?
Yes. Attach up to 10 files per message, 10 MB per file and 20 MB in total. PDF and Word files can be up to 2 MB each, and up to 80,000 characters of their text are read. Images (JPEG, PNG, WebP, GIF) and text or code files work too.
Can the AI give me a real PDF, Word doc, PowerPoint or Excel file?
Yes. Ask for one and the reply carries the file with Open and Download. Text and code formats such as csv, html, json, svg and sql come back as real files too. Formats a text model cannot honestly produce, such as exe, zip, mp3, mp4, png and epub, are refused rather than faked.
Can several AI models answer the same question?
Yes. Consensus sends your question to three models at once, and a judge writes one verdict. An optional Round 2 adds rebuttals, priced separately. In Debate, three models argue For, Against and Neutral over 1 to 5 rounds before a judge rules. Both come in a Standard and a Pro lineup.
How much does SynergyAI chat cost?
One account covers every studio: top up as you go, or choose a monthly Pro or Studio plan. A chat reply costs what it uses, shown under it once it lands, and a Max per reply you set means none costs more. A reply that fails costs nothing. Consensus and Debate are priced per run, with extra rounds priced separately, and Voice Mode by the minute.
Can I practise a job interview with AI?
Yes. Coach mode has 18 role-play scenarios, including tech, PM and HR interviews, consulting cases, UPSC, VC pitches and IELTS speaking. Pick easy, realistic or hard. You are scored live out of 100, get a per-dimension feedback report, and can do it out loud in Voice Mode.
Does SynergyAI chat remember my past conversations?
Yes. Each conversation keeps its own memory: older messages are recalled when relevant and the rest is summarised, with no change to your price. History syncs across your devices. In Forget and Burn mode nothing is written to the server at all.
Open a chat and pick your model
Start free on GPT-4o Mini or Gemini Flash Lite, 500 calls a day each in main chat without web search, and move up a tier when a question needs more.