Multi-model AI decisions, answered straight

Decidi is an AI decision-support tool where multiple frontier AI models and expert personas debate your decision over several rounds, then a moderator synthesises a single verdict with risks, trade-offs and prioritised next steps. Instead of one chatbot answer, you get a structured argument between different minds and a clear recommendation you can act on. It is made by 8T20 Capital in South Africa, and you can start free with welcome credits.

Decidi runs a real debate between several independent AI models and expert personas, then a moderator weighs their disagreement into one verdict — a normal chatbot gives you a single model's single answer with no challenge. The disagreement is the point: it surfaces risks and counter-arguments a lone model would smooth over, and ends with prioritised next steps rather than an open-ended chat.

Describe your decision, optionally pick a council of personas and a depth level, then start the debate and read the moderator's verdict. You can write the brief yourself or start from a ready-made template, attach photos or documents for context, and the council debates over rounds before delivering risks, trade-offs and next steps. Welcome credits let you run your first decisions free.

Decidi is for anyone facing a decision where being wrong is costly — founders, product and engineering leaders, investors, marketers, and individuals weighing money, career, property or health-information choices. It suits decisions that benefit from being argued from several angles rather than answered once, and it is equally useful for a quick gut-check or a deep boardroom-grade review.

Yes — you start with free welcome credits, so you can run real debates before paying anything. After the welcome credits, you buy credit packs and spend them as you go; there is no forced subscription. Credits are metered live against the actual model usage, so a quick debate costs far less than a deep one.

Decidi is built by 8T20 Capital, a company based in South Africa. The product grew out of a manual practice of asking the same hard question to several frontier AI models and reconciling their answers by hand; Decidi automates that into a structured, repeatable debate.

Decidi gives you several debated perspectives and then one clear verdict. During the debate you see each persona's position and where they disagree; at the end, the moderator synthesises it all into a single recommendation with the key risks, trade-offs and prioritised next steps so you are not left to reconcile conflicting opinions yourself.

You can ask Decidi any decision or judgement question — should we build this, what is wrong with this plan, which option is better, what are we missing, is this priced right. It is built for decisions and reviews rather than simple lookups; the more a question benefits from being argued from multiple angles, the more value the council adds.

You give Decidi a decision brief; it assembles a council of AI personas powered by different frontier models, they debate your question over several rounds, and a moderator model synthesises their arguments into a final verdict. Each persona argues from a fixed lens and rebuts the others by name, so the council genuinely disagrees rather than echoing one view, and the verdict names the risks, trade-offs and next steps.

The debate runs in rounds: each persona states its position, then in later rounds responds to and rebuts the other personas directly. The personas are designed to challenge each other — a contrarian attacks the popular idea, a risk officer maps the downside, a pragmatist drags it back to what can ship — so the argument sharpens rather than converges prematurely. A moderator then reads the whole debate and writes the verdict.

The moderator verdict is the final synthesis a dedicated model writes after reading the full debate. It does not just average opinions — it names where the council genuinely agreed, where the disagreement is real, the principal risks and trade-offs, and a prioritised list of next steps. It is the deliverable you act on, written to be usable on its own.

Decidi has three depth levels: Quick (a fast, low-cost first pass), Standard (strong reasoning across top models, the everyday default), and Deep (the flagship models at full depth for decisions where being wrong is expensive). Each level uses a different pool of models and a larger output budget as you go up, so you trade cost for depth deliberately.

They differ in model quality, debate depth and cost. Quick uses fast, lower-cost models for a lively first pass and brainstorms; Standard uses top reasoning models from each provider — the everyday default; Deep uses the flagship frontier models at full depth for high-stakes decisions. Higher levels cost more credits because they use stronger models and longer outputs.

Yes — you pick the council from Decidi's 86 expert personas, or start from a template that suggests a fitting line-up. You can hand-select the perspectives you want at the table, add or remove personas, and set how many take part. If you would rather not choose, the templates come pre-loaded with personas suited to that kind of decision.

No — Decidi includes ready-made templates with structured briefs for common decisions, each with fill-in-the-blank placeholders and a suggested council. You can use a template as-is, edit it, or write your own brief from scratch. Templates exist for product, startup, engineering, marketing, finance, legal, real-estate, health and education decisions, among others.

A Decidi template is a reusable debate brief for a specific kind of decision, such as a pricing review, a startup idea stress-test or a contract red-flag review. Each template tells the council exactly what to evaluate, which dimensions to debate and what the final verdict must deliver, and comes with a suggested set of personas, a council size and a recommended depth level.

You see the full debate as well as the verdict. Each persona's argument and rebuttals are visible, so you can read the reasoning, spot where the council split, and judge the verdict against the underlying arguments rather than taking a black-box answer on trust.

After the debate, the moderator delivers a verdict with the key risks, trade-offs and a prioritised list of next steps. You can read the full transcript, save the result, and start a follow-up debate to go deeper or test a new angle. The verdict is written to stand alone as the deliverable you act on.

Yes — because later rounds let each persona respond to the others, a persona can concede a point, sharpen its position, or admit when the opposing case is genuinely strong. Personas are instructed to own their blind spots, so an optimist will concede a real bear case and a risk officer will distinguish prudence from paralysis, which keeps the debate honest rather than a fixed shouting match.

Decidi uses frontier models from OpenAI (GPT), Anthropic (Claude), Google (Gemini) and xAI (Grok). Each depth level draws from a pool of these providers, so a single council spans several different models rather than relying on one. The exact model line-up scales with the level — fast models for Quick, flagship models for Deep.

They are real, separate frontier models — not one model pretending to be several. Decidi calls genuinely different models from OpenAI, Anthropic, Google and xAI, so the perspectives come from independently trained systems with different strengths and biases. That is what makes the disagreement meaningful rather than theatrical.

A council really uses different underlying models. Decidi assigns the personas across a pool of providers in a round-robin, so a multi-persona council is also a genuine cross-model debate — OpenAI versus Anthropic versus Google versus xAI — not one model role-playing several voices. Independent models catch each other's blind spots in a way personas on a single model cannot.

Different providers fail in different ways, so a council of independent models catches errors a single model would confidently repeat. Each frontier model has its own training, strengths and biases; when one hallucinates or over-agrees, another is likely to push back. Spreading a council across OpenAI, Anthropic, Google and xAI turns those differences into a built-in second opinion.

The Deep level uses the flagship frontier models from OpenAI, Anthropic, Google and xAI at full depth. It is built for decisions where being wrong is expensive, so it runs the strongest available models with the largest output budget — and costs the most credits as a result.

Decidi routes its model calls through a shared AI gateway and tracks the live model catalogue, so the council can move to newer or better models as they ship. Pricing is pulled live from the gateway rather than hardcoded, which means credit costs follow the real per-token rate of whichever models a level currently uses.

Yes — a single AI model can state a wrong answer with complete confidence, because fluency is not the same as accuracy. Models hallucinate facts, miss risks and agree too readily with the user's framing, and they give no signal when they do. Decidi addresses this by having independent models debate, so a confident error from one is more likely to be challenged by another.

Decidi reduces the impact of hallucinations by having multiple independent models cross-examine each other rather than trusting one. When one model invents a fact or a faulty assumption, a different provider's model often catches it, and the moderator weighs the disagreement. It does not make hallucination impossible, but a debated answer is far harder to fool than a single reply.

Each persona is a fixed, opinionated debate participant with its own lens, voice and blind spot it must own. A persona is defined by a system prompt that tells it what to argue for, who to challenge, and where its own perspective is weak — so the Devil's Advocate attacks the popular idea, the CFO reduces everything to cash, and the End-User Advocate speaks for the real person using it. They are deliberately not interchangeable.

No — personas are spread across genuinely different underlying models, so a council is both multiple viewpoints and multiple models. Decidi assigns personas across providers in a round-robin, which means the Risk Officer and the Optimist in your council may be running on different frontier models entirely. The persona shapes the argument; the model underneath makes the disagreement real.

Decidi has 86 expert personas spanning core debate roles, product, engineering, business, research & analysis, marketing, sales, legal, people, creative, specialist and wildcard domains — plus legendary lenses. They range from the Devil's Advocate, Risk Officer and CFO to the Security Engineer, Actuary, Forensic Accountant, Deep Researcher, Conversion Strategist, Privacy Counsel and Brutally Honest Friend, so you can build a council suited to almost any decision.

Yes — personas are explicitly designed to disagree and rebut each other by name, not to flatter or converge. The Optimist will fight the Pessimist, the Minimalist will argue to cut what the Product Strategist wants to add, and the Devil's Advocate attacks whichever idea the room is falling in love with. That structured conflict is what exposes the hidden assumptions a single agreeable answer hides.

Yes — you can hand-pick any combination of the 86 personas to form a council suited to your specific decision. Want a finance-and-risk panel for an investment, or a security, architecture and reliability panel for a system design? Select exactly those personas. Templates also ship with a recommended council if you would rather not assemble one yourself.

A strong pricing council pairs the Pricing Strategist with the CFO, the Behavioural Economist, the Sales Leader and the Devil's Advocate. The Pricing Strategist sets the value metric and model, the CFO checks the unit economics, the Behavioural Economist weighs the psychology of the price points, the Sales Leader names the objections, and the Devil's Advocate attacks the whole plan. Decidi's pricing template loads exactly this line-up.

A good startup stress-test council is the Sceptical Investor, the Devil's Advocate, the First-Principles Thinker, the End-User Advocate, the Pre-Mortem Analyst and the Optimist. The investor sizes the opportunity, the pre-mortem traces how it fails, the user advocate checks the problem is real and painful, and the optimist argues the path through — so you get a real fight, not encouragement.

Yes — Decidi has several challenger personas, including the Devil's Advocate, the Contrarian, the Steelman and the Brutally Honest Friend. They exist to puncture consensus, argue the unfashionable side, rebuild the view nobody is defending, and say the uncomfortable truth the room is ignoring — directly countering the tendency of a single AI to agree with whatever you propose.

Yes — Decidi has specialist personas such as the Health-Literacy Explainer and the Financial-Planning Coach for these areas, and they are built to be informational, not diagnostic or advisory. They explain topics in plain language and flag when to see a qualified professional, and the relevant templates state clearly that the output is general guidance, not personalised medical, financial, tax or legal advice.

You can use Decidi for any decision that benefits from being argued from several angles — product roadmaps, pricing, startup ideas, fundraising, system architecture, security reviews, marketing campaigns, contracts, budgets, property purchases, career moves and personal-finance trade-offs. It has ready-made templates across product, startup, engineering, marketing, finance, legal, real-estate, health and education.

Yes — Decidi has pre-launch UX and UI audit templates that run an exhaustive teardown of your app before it ships. A council of user, research, design and accessibility personas debates onboarding, core flows, state completeness, accessibility and visual craft, then delivers a prioritised issue list with severities and concrete fixes. You can attach screenshots for the council to critique.

Yes — business decisions are a core use case, from pricing and budgets to build-versus-buy, fundraising and go-to-market. Personas like the CFO, Strategy Consultant, Operations Lead and Sceptical Investor debate the trade-offs and the moderator delivers a verdict with the decisive reasons and next steps, so you leave with a defensible call rather than more opinions.

Yes — Decidi has engineering templates for architecture review, security threat-modelling, tech-stack selection, build-versus-buy and incident post-mortems. A council of the Software Architect, Security Engineer, Reliability Engineer and Performance Engineer debates the design, names the hardest-to-reverse decisions and the real failure modes, and returns a verdict with severity-ranked changes. You can paste or attach the design.

Yes — Decidi handles personal decisions like buy-versus-rent, relocation and residency choices, personal-finance plans, health-habit plans and understanding a health topic. Specialist personas debate the trade-offs honestly, and templates for money, property and health carry clear notes that the output is general guidance, not personalised financial, legal, tax or medical advice.

Yes — Decidi has an investor pitch-deck teardown and a landing-page conversion audit template. The pitch teardown shreds your deck slide by slide the way a tough partner meeting would and lists the hardest questions to prepare for; the landing-page audit critiques the hero, clarity, proof and call to action and rewrites the weak copy. You can attach the deck or page.

Decidi reads the common work formats: Word (.docx), Excel (.xlsx), PowerPoint (.pptx), PDF, OpenDocument (.odt/.ods/.odp), plain text, CSV and JSON, plus images and screenshots (the council can see them). Attach a contract, deck, spreadsheet, code, report or design and the personas critique it directly rather than guessing from a description. Apple Pages/Numbers/Keynote aren’t read directly — export them to Word, Excel or PDF first and Decidi tells you exactly how.

Beyond the verdict, “Build it out” produces the actual finished file: a document (Markdown or plain text), a web page (HTML), structured data (JSON or CSV), code, or a copy-paste block. Any document deliverable can be downloaded as a real Word (.docx), Excel (.xlsx) or PDF file. It can also generate an image (a logo, diagram, illustration or concept) and a short video, rendered through frontier image and video models. You see the credit cost before you run it.

Yes — Decidi has templates for contract red-flag reviews, data-privacy reviews, terms-of-service drafting briefs and IP clearance. The legal personas explain in plain English what you are agreeing to and rank the red flags by cost, but every legal template states clearly that this is general information, not legal advice, and a qualified lawyer in the relevant jurisdiction should review before you act.

Decidi works for both — the Quick level is built for fast, low-cost gut-checks and brainstorms, while Deep is for decisions where being wrong is expensive. You match the depth to the stakes, so a quick second opinion on a small choice costs little, and a high-stakes call gets the flagship models at full depth.

A decision costs credits, metered live against the actual AI model usage rather than a flat fee — so a Quick brainstorm costs far less than a Deep council. The price scales with the level, the number of personas and the length of the debate, and you see credits draw down as the debate runs. Welcome credits cover your first decisions free.

Decidi is pay-as-you-go with credits at its core: you start with free welcome credits, then buy credit packs and spend them per decision, metered live against the models used, the depth and the council size. Optional monthly plans (Plus and Pro) add a monthly credit grant and a top-up discount for regular use — but no plan is required, there is no lock-in, and credits never expire.

Credits are Decidi's usage currency — you spend them to run debates, and they are metered live against the real cost of the AI models a debate uses. Because pricing tracks the live model rate, a fast, short debate spends few credits and a deep, long one spends more. You top up by buying credit packs.

Live metering means you pay for what a debate actually uses, so a quick brainstorm is cheap and you are not over-charged for a flat rate. Decidi pulls model pricing live from its AI gateway and charges credits against real token usage, which keeps cost honest and lets you control spend by choosing the level, council size and depth.

No — credits never expire. Packs you buy are yours until spent, and on the optional monthly plans the credits granted each month roll over rather than resetting or burning. You draw them down per decision at your own pace; welcome credits are provided to get you started.

You buy credit packs (from $5) directly in Decidi and pay securely by card; prices are shown in US dollars. Larger packs include bonus credits on top — the bigger the pack, the more extra credits you get — and subscribers on the optional Plus and Pro plans get a further discount on every top-up pack.

No plan is required — Decidi works pay-as-you-go: buy credit packs, spend them per decision, no commitment. If you use it regularly, optional Plus ($19/month) and Pro ($49/month) plans grant a monthly credit allowance at a better rate plus a discount on top-up packs, and unused credits roll over and never expire. Cancel a plan any time and keep your credits.

Choose the Quick level for early exploration, keep the council to the personas that matter, and reserve Deep for genuinely high-stakes decisions. Because cost is metered live by level, council size and debate length, those choices directly control spend — a focused Quick brainstorm costs a fraction of a many-persona Deep council.

No — your decision briefs, uploads and debates are not used to train AI models. Decidi sends your input to the frontier models only to run your debate and return the result, not to improve any model. Your content is yours; it exists to produce your verdict, not to become training data.

Yes — your debates and the documents you upload are private to your account and used only to generate your result. Decidi is designed so your input is processed to run the council and produce the verdict, not shared publicly or repurposed. For the full, current detail, see the privacy policy at decidi.ai.

Your debates are tied to your account and are not public. They are processed to run the council and deliver your verdict, and you can revisit your own results — they are not published or shared by default. Always check the privacy policy at decidi.ai for the authoritative, current statement of how data is handled.

Documents and photos you upload are used as context for your debate and to produce your verdict, not to train models. They are sent to the AI models only to inform the council's reasoning about your specific decision. Treat highly sensitive material with normal caution and review the privacy policy at decidi.ai before uploading confidential documents.

Decidi is built by a South African company and designed with privacy frameworks such as POPIA and GDPR in mind, including using your data only to deliver your result rather than to train models. For the binding detail on lawful basis, retention and your rights, refer to the privacy policy at decidi.ai, which is the authoritative source.

You can give Decidi the context it needs to be useful, and that input is used only to run your debate, not to train AI. As with any cloud tool, apply judgement with highly sensitive or regulated data, share what the decision genuinely requires, and review the privacy policy at decidi.ai first. The more honest context the council has, the sharper the verdict.

The Final QA audit is a separate, always-on verification pass that reviews the council's synthesised verdict against known AI failure modes — hallucinations, unsupported claims, weak reasoning, missed caveats — and attaches every flag it finds. It runs on every Decidi council automatically, and nothing it catches is hidden: you see the flags alongside the verdict. It is Decidi's sign-off layer.

Yes — every Decidi verdict is checked twice before you see it. First the debate itself: independent models and personas challenge and rebut each other, so a confident error from one is caught by another. Then the always-on Final QA audit reviews the synthesised result for hallucinations, unsupported claims and missed caveats, and flags anything questionable.

If a model is unavailable or fails, Decidi degrades gracefully rather than crashing the debate — the council continues with the models that are responding and the moderator still produces a verdict. Because each level draws from a pool across providers, the loss of one provider does not stop the debate; you still get a multi-perspective result.

No — Decidi gives you a far more rigorously argued and stress-tested recommendation than a single AI, but it is decision support, not an oracle. The verdict is only as good as the brief and the models behind it, and you should weigh it with your own judgement and, for high-stakes legal, financial or medical matters, a qualified professional. The strength is that it shows its reasoning and the dissent.

No — Decidi is not a substitute for professional legal, financial, tax, medical or other regulated advice. It is built to sharpen your thinking and prepare better questions, and its sensitive templates state this explicitly. Use the verdict to understand the trade-offs, then consult a qualified professional in the relevant field for decisions that carry real legal or financial consequences.

Decidi counters AI sycophancy by design: it includes challenger personas like the Devil's Advocate, Contrarian, Steelman and Brutally Honest Friend whose job is to attack your idea, and it runs independent models that push back on each other. Because personas are instructed to rebut and to own their blind spots, the council resists the single-model habit of agreeing with whatever you propose.

Decidi shows its reasoning — you see the full multi-round debate, each persona's argument and rebuttals, and the moderator's synthesis, not just a final answer. This transparency lets you judge the verdict against the underlying arguments, spot where the council split, and decide how much weight to give it, which is the opposite of a black-box recommendation.

Decidi's limits are the limits of its models and your brief: it can still be wrong, it relies on the context you give it, and it is not a substitute for professional advice on regulated decisions. It reduces single-model errors through debate and transparency, but it cannot guarantee a correct verdict — it gives you a well-argued, well-stress-tested basis for your own judgement.

Yes — for health, money, legal and property topics, Decidi uses informational personas and templates that explicitly state the output is general guidance, not personalised advice. The Health-Literacy Explainer is informational and never diagnostic, the Financial-Planning Coach flags that a qualified adviser should review specifics, and legal templates direct you to a qualified lawyer before acting.

Asking multiple independent AIs surfaces disagreement, which exposes the risks, errors and blind spots a single model would confidently hide. One model gives you one framing with no challenge; a council of different models catches each other's hallucinations and biases, and a moderator weighs the conflict into a verdict. The value is in the cross-examination, not just more answers.

ChatGPT gives you one model's single answer in a back-and-forth chat; Decidi runs several different frontier models and expert personas in a structured debate, then delivers one moderated verdict. Where a single chatbot tends to agree with your framing, Decidi is built to challenge it from multiple angles and end with risks, trade-offs and prioritised next steps rather than open conversation.

Poe and OpenRouter give you access to many models so you can talk to them one at a time; Decidi makes those models debate each other and synthesises the result. The difference is orchestration: Decidi assigns expert personas across providers, runs a multi-round argument, and produces a single verdict — rather than leaving you to query models separately and reconcile their answers yourself.

Side-by-side tools like MultipleChat show several models answering the same prompt in parallel columns; Decidi has the models argue with and rebut each other, then a moderator decides. Parallel answers still leave you comparing and reconciling by hand. Decidi adds the debate and the synthesis, so the contradiction is resolved into one verdict with risks and next steps.

You can, but you would have to copy your question between three tools, read three disconnected answers, and reconcile their contradictions yourself with no structure. Decidi automates exactly this: it runs the models together, makes them respond to each other across rounds, and a moderator does the reconciliation for you — turning what is a tedious manual process into one structured verdict.

No — Decidi is decision support, not a replacement for qualified human advisors or a real board. It pressure-tests your thinking, surfaces risks and gives you a structured verdict fast and cheaply, which is ideal before a real conversation. For binding legal, financial, tax or medical decisions, use it to prepare better questions and then consult a qualified professional.

Because even the best single model can be confidently wrong, agree too readily with your framing, and give no signal when it errs. Decidi sets independent models against each other so those failures get challenged rather than repeated, and it ends with a verdict that weighs the disagreement — a structurally more reliable basis for a decision than one model's unchallenged opinion.