Canonity logo
Agents Orchestration Frameworks · Workflow Automation

Canonity

Canonity is an Italian no-code platform that orchestrates nine large language model providers in one console. Chat across models, switch engine mid-conversation, then chain prompts into visual, repeatable workflows that produce documents, images and scraped data.

Active GDPR compliant Free plan Freemium API available 18+ Verified by Guidaio
Overview

What is Canonity?

Canonity is a web platform for orchestrating several large language models at once, published by Canonity SRL, an innovative startup based in Palermo, Italy. Its argument is stated plainly on the technology page: no generalist model is best at everything, so instead of asking one model to do the whole job you assemble what the publisher calls a council of experts, calling each engine at the moment it is strongest. The design is openly credited to Andrej Karpathy's open source llm-council project, in which several models answer, critique one another, and a final model synthesises the result.

The product comes in three levels of ambition. The Chatbox is a multi-model chat where you can change engine mid-conversation, send the same prompt to two or three models and promote the best answer into the main thread, or let auto-routing choose for you; every reply stays labelled with the model that produced it, and tokens and cost are displayed as you go. The Chain editor links prompts in sequence, each step feeding the next. The visual Workflow editor is the no-code heart of the product: you lay out Steps and Buffers, assign a model to each step, and replay the whole process as often as you need. A tool called the Canonizer builds a working workflow from a spoken description of the task.

Nine providers are integrated, namely OpenAI, Anthropic, Google, xAI, Mistral, DeepSeek, Moonshot AI, Alibaba and Zhipu AI, and the home page claims twelve model families across more than thirty versions. Workflows can also produce documents and images, accept uploaded files, scrape the web and call third-party APIs. Storage runs from 5 to 200 GB depending on plan, and content is encrypted at rest with a distinct key for every file.

The company is young and openly self-funded. It was incorporated on 6 May 2026, has taken no investment, and has filed a patent application on its workflow editing and execution system. A u-prompt marketplace, where users would sell the execution of a workflow without revealing the source prompt, is announced for January 2027. The interface is available in Italian, English, French, German and Spanish, and the product is web-only.

What it does

  • Chat with nine LLM providers from a single console and switch engine in the middle of a conversation
  • Send the same prompt to two or three models, compare the answers and promote the best one into the main thread
  • Break a task into chained steps, each running the model best suited to it
  • Build a visual, no-code workflow and replay it as many times as needed
  • Generate a ready-to-run workflow from a plain description of the task using the Canonizer
  • Produce documents and images, and upload your own files into a workflow
  • Add web scraping and third-party API calls as steps inside a workflow
Audience

When to use Canonity / When not to

A quick filter to help you decide if Canonity is the right fit.

When to use Canonity

  • Marketing and content teams that want several models writing, reviewing and refining the same piece without copy-pasting between tools
  • Copywriters, strategists and consultants who write no code but want to turn a proven method into a repeatable process
  • Small businesses and independent professionals looking for one paid console instead of separate subscriptions to several assistants
  • European organisations that need their orchestration layer hosted inside the EEA and want each provider's GDPR risk stated before they send a prompt
  • Builders who intend to package and sell the execution of a workflow through the announced u-prompt marketplace

When not to use Canonity

  • Anyone needing an answer that can be relied upon as professional advice: the terms explicitly exclude medical, legal and financial use
  • Teams handling sensitive material, since the terms forbid entering passwords, card numbers or medical data into prompts
  • Developers who need documented API access today: the API is billed on the top plan but no reference is published and the roadmap dates it October 2026
  • Buyers who require a signed Data Processing Agreement or a recognised security certification before onboarding a vendor
  • Mobile-first users, as Canonity is web-only with no iOS, Android, desktop or browser-extension version
Get started

How to use Canonity

A typical end-to-end flow, from setup to results.

  1. Open the home page and type straight into the chatbox: no account, no installation and no configuration are needed, within an anonymous quota
  2. Pick a model from the selector, then switch to another one mid-conversation; each reply stays labelled with the engine that produced it
  3. Send the same prompt to two or three models side by side and promote the answer you prefer into the main thread
  4. Create a free account to keep your history, upload files and use the storage included with your plan
  5. Move to the Chain editor to link prompts in sequence, feeding each step's output into the next
  6. Describe your task to the Canonizer and let it assemble a ready-to-run workflow for you
  7. Open the visual Workflow editor to lay out Steps and Buffers, choosing a model per step or leaving auto-routing to decide
  8. Add web scraping or third-party API steps where a workflow needs external data, supplying your own credentials
  9. Watch the token count and cost shown on each reply to keep spending visible
  10. Subscribe from the subscriptions page when the free quota runs short: billing is monthly through Stripe, and cancellation is handled from the active plan
Quick read

Pros & Cons

Pros

  • Nine model providers in one console, including European and Chinese engines rarely offered side by side
  • Model switching mid-conversation, with every reply attributed to the engine that produced it
  • Orchestration infrastructure hosted entirely in the EEA at OVH, across France, Germany and Poland
  • Unusually candid privacy page that rates each provider's GDPR risk instead of avoiding the question
  • Tokens and cost shown per reply, so spending stays visible rather than arriving at the end of the month
  • A genuinely permanent free plan, and a chatbox that works without signing up at all
  • Detailed, checkable security page covering the algorithm, key derivation and deletion behaviour

Cons

  • Three incompatible price lists coexist on the site: the home page, the pricing page and the checkout page each show different plan names and different amounts
  • Documentation and tutorials are advertised but empty, both pages reading only Coming soon
  • API and MCP access is billed on the top plan while no reference is published, and the roadmap places the API in October 2026
  • The home page shows a Used by teams at banner whose twelve company names are hard-coded in the JavaScript bundle under a variable literally called fakeLogos
  • No Data Processing Agreement is offered, and the publisher claims no security certification of its own
  • Payments are non-refundable with no pro rata on cancellation, and deletion is immediate and irreversible with no recycle bin
  • A very young, self-funded company incorporated on 6 May 2026, with parts of the English interface still showing Italian labels
Pricing

Pricing & Plans

A permanent free plan is available, offering 930 monthly credits, a ceiling of 300 per day and 5 GB of storage on Economy-tier models. The lowest paid entry point on the page that actually processes payment is the Base plan at EUR 14.99 per month, VAT included, billed monthly through Stripe. Prospective subscribers should note that the site publishes two further, divergent price lists elsewhere; the subscriptions page is the one that governs what is charged.

Free, EUR 0
  • 930 monthly credits
  • 300 per day
  • 5 GB of storage
  • Economy-tier models
  • chatbox only
Plus, EUR 35.00 per month VAT included
  • 49
  • 000 monthly credits
  • 16
  • 000 per day
  • 100 GB of storage
  • Premium-tier models
  • Advisor
  • Canonizer twice a day
Pro, EUR 50.00 per month VAT included
  • 70
  • 000 monthly credits
  • 30
  • 000 per day
  • 200 GB of storage
  • Premium tier plus LLM Professional
  • unlimited Canonizer
  • unlimited workflows
Prices and plans listed above may evolve. Always check the official pricing page before subscribing.
Trust & Privacy

Data, GDPR & hosting

A consolidated view of how Canonity handles your data.

GDPR overview

GDPR treatment is explicit rather than implied. The privacy page carries a structured summary naming Canonity SRL, Via Valdemone 31 in Palermo, as controller, with privacy@canonity.com for requests and a thirty-day response commitment. Rights under Articles 15 to 22 are listed individually, and the Italian supervisory authority, the Garante per la Protezione dei Dati Personali at Piazza di Monte Citorio 121 in Rome, is named for complaints. Infrastructure is stated to sit in the EEA, and transfers outside the European Union are framed as a consequence of the model provider you select. The minimum age is 18. Two gaps deserve mention: no Data Processing Agreement is offered or described as available on request, and the publisher claims no security certification of its own. Being established in Italy, it needs no Article 27 representative, and none is named.

Who owns the data?

Canonity presents itself as a technical intermediary rather than an owner of content. Its terms make you the party who decides which provider is used and what is sent to it, and you carry responsibility for copyright and GDPR compliance on those choices. For prompts and outputs routed to a model, Canonity states that it acts as your data processor. For its own account records, including email, name, hashed password, billing data and technical logs, Canonity SRL is the controller, reachable at privacy@canonity.com. Stored content is encrypted at rest with a key derived per user and per file path, and the master key sits on backend infrastructure kept separate from the application servers.

Reuse rights

Canonity claims no ownership over what you produce with it, and nothing in the terms requires you to ask permission before reusing your own output. From the Pro plan onwards you are explicitly invited to sell the execution of a workflow through the u-prompt marketplace without exposing the source prompt. What leaves the platform is narrow: only prompt text and technical parameters reach the model provider, while your IP address, email, payment details and device information do not. Canonity states that it has no model of its own, does not train, does not sell data, does not run behavioural advertising and does not monitor prompt content. The provider you select, however, applies its own rules, and the privacy page records them: DeepSeek, Moonshot AI and Zhipu AI train on submitted data, xAI may use it for improvement, and Anthropic offers an opt-out. Choosing a non-EU provider is presented as accepting that transfer.

Data retention & training

Retention summary
Retention is stated by category. Account data is kept for the duration of the service plus 30 days. Payment records are held for seven years to meet Italian tax obligations. Security logs last 90 days. Prompts and generated content are not stored beyond the execution itself. After you delete your account, or 30 days after termination, Canonity says it removes all personal data and user content, keeping only what the law requires for the minimum period. Individual deletions are physical: removing an execution, a chat or an uploaded file unlinks it from the filesystem immediately, with no recycle bin, no grace period and no shadow copy, and neither the user nor Canonity can recover it afterwards. The privacy page is dated 1 September 2026 in its header, though its footer still reads version 2.1 of 29 June 2026.
Trains on customer data
No
Subprocessors disclosed
Yes
GDPR contact

Hosting summary

Canonity states that its own infrastructure sits inside the European Economic Area, at OVH, across France, Germany and Poland; the site's public address resolves to an OVH range in Frankfurt, Germany. Content stored on the platform, meaning chats, executions and uploaded files, is encrypted at rest with a key derived for each individual file from the user identifier and the file path, while the master key is held on backend infrastructure separate from the application servers and never exposed to the browser. Data leaves that perimeter as soon as a prompt is sent to a model, and hosting then follows the provider chosen, which the privacy page maps explicitly: Mistral in France, OpenAI, Anthropic and xAI in the United States, Google in the United States or the European Union, Alibaba in Singapore or China, and DeepSeek, Moonshot AI and Zhipu AI in China. The publisher warns in its own words that data held under Chinese law may be subject to government access. No paid data-residency option exists: the wording appears only on a home page plan grid that cannot be purchased.

Hosting countries
🇫🇷 France🇩🇩 Germany🇵🇱 Poland
Hosting regions
EEA
Watch-outs

Things to keep in mind

Risks and trade-offs to weigh before adopting Canonity.

  • Verify the price on the subscriptions page before paying: the home page and the marketing pricing page advertise different plans and different amounts, and only the subscriptions page drives the charge
  • The Used by teams at banner names twelve companies that are hard-coded placeholders in the site's JavaScript, not customers, and must never be treated as references
  • API and MCP access is charged on the top plan while no documentation exists and the roadmap places the feature in October 2026, so confirm availability before subscribing for that reason
  • Training policy follows the provider you pick rather than Canonity: sending a prompt to DeepSeek, Moonshot AI or Zhipu AI means accepting that it may be used for training and held under Chinese law
  • Deletion cannot be undone, with no recycle bin, no grace period and no backup copy, so download anything you want to keep before removing it
  • The ISO 27001 certification mentioned on the roadmap belongs to the MIIA infrastructure operated by Fastweb and announced for September 2026; it is not a certification held by Canonity
  • Convenience can erode judgement: chaining models makes output arrive faster and look more authoritative, while the terms are explicit that nothing is verified and that it must not be treated as professional advice
Setup

Setup & Integrations

Technical difficulty

Very low to start, moderate to master. The home page chatbox answers immediately with no account, no installation and no configuration, and the product runs entirely in the browser. The real learning curve is the workflow editor, where you have to think in Steps, Buffers and per-step model choice; the Canonizer softens it by generating a workflow from a plain description. Two things make it harder than it should be: no documentation or tutorials are published yet, and third-party API steps require you to manage your own credentials and respect the target API's terms.

Deployment

Web appAPI

Integrations

OpenAI Anthropic Google XAI Mistral DeepSeek Moonshot AI Alibaba Zhipu AI Stripe

Supported languages

ItalianEnglishFrenchGermanSpanish
Company

Behind Canonity

Company name
Canonity SRL
Founded
06/05/2026
Country of origin
🇮🇹 Italy
Headquarters
Via Valdemone 31, 90144 Palermo (PA), Italy
UBO
Stefano Candela
UBO country
🇮🇹 Italy
Domain registrar country
🇫🇷 France
Legal contact
Support contact

Social

Official links

Resources

All the official URLs gathered for verification and reference.

Compare

Alternatives

Tools that compete with or complement Canonity.

C ChatGPT
FAQ

Frequently asked questions

How is Canonity different from using ChatGPT directly?
ChatGPT answers one prompt at a time. Canonity chains several steps, gives each step the model best suited to it, keeps memory between steps in Buffers, and lets you design a process once and replay it whenever you need it.
Which model providers can I reach through Canonity?
Nine are integrated: OpenAI, Anthropic, Google, xAI, Mistral, DeepSeek, Moonshot AI, Alibaba and Zhipu AI. The home page claims twelve model families across more than thirty versions.
Do I need an account to try it?
No. The chatbox on the home page works without signing up, within an anonymous quota. An account becomes necessary to keep your history, upload files and use your storage.
What does it cost?
There is a permanent free plan. The cheapest paid plan on the page that processes payment is Base, at EUR 14.99 per month with VAT included. Be aware that the home page and the marketing pricing page each display a different price list.
Where is my data hosted?
Canonity states that its own infrastructure sits in the EEA, at OVH in France, Germany and Poland. Prompts you send to a model leave that perimeter for wherever the chosen provider operates.
Does Canonity train models on my data?
Canonity states that it has no model of its own and does not train. The provider you select applies its own policy, and Canonity's privacy page notes that DeepSeek, Moonshot AI and Zhipu AI do train on submitted data.
Is there an API?
API and MCP access is listed and billed on the Pro plan, and the terms include a section for it, but no reference documentation is published and the roadmap dates the API to October 2026. Treat availability as unconfirmed.
Is there a mobile app?
No. Canonity is web-only, with no iOS, Android, desktop or browser-extension version.
What happens when I delete something?
Deletion is physical and immediate. There is no recycle bin, no grace period and no shadow copy, and neither you nor Canonity can recover the file afterwards. Download anything you want to keep before removing it.
What is the minimum age to use Canonity?
Eighteen. The terms of service require users to be at least 18 years of age.
Conclusion

Should you pick Canonity?

Canonity delivers on its central idea. The multi-model thesis is not merely marketing: the chatbox really does let you change engine mid-conversation and keeps track of which model said what, the chain and workflow editors make the sequencing concrete, and the per-reply cost display answers the question most multi-model tools leave open. For a European buyer the differentiators are real. The orchestration layer is hosted inside the EEA at OVH, the encryption design is documented in checkable detail, and the privacy page grades each model provider's GDPR risk instead of glossing over it. Very few competitors publish that table.

What lets it down is the site around the product. Three different price lists coexist, with different plan names and different amounts, and only the subscriptions page reflects what is actually charged. Documentation and tutorials are advertised and empty. API and MCP access is billed on the top plan while no reference exists and the roadmap places the feature a month after this review. The home page displays a customer banner whose twelve names are hard-coded placeholders in the JavaScript. None of this makes the tool worse, but it makes the publisher harder to read.

The sensible approach is the one the product itself invites: use the free chatbox without an account, run a real task through it, then check the subscriptions page rather than any other page before paying. This is a very young, self-funded company, incorporated in May 2026, so treat forward-looking features as promises rather than as inventory.