Curiso
Curiso is an infinite node canvas for working with large language models. Chat, note and research nodes pass context between each other, connecting twelve cloud providers or five local inference engines through your own API keys.
What is Curiso?
Curiso is an infinite canvas for thinking with large language models. Rather than a single linear chat thread, you place nodes on an unlimited board and wire them together, so context flows from one node to the next without being restated at every turn. The site describes three node types: Chat Nodes for talking to models and passing context downstream, Notes Nodes for capturing ideas and feeding them into a chat, and Deep Research for working through complex subjects.
The tool is deliberately bring-your-own-key. Curiso does not resell inference; you connect your own accounts and pay each provider directly. The homepage grid lists twelve remote providers — OpenAI, Anthropic, Google AI, OpenRouter, xAI, Groq, W.ai, Inference.net, Deepseek, Fal and Alibaba, plus a slot for others. The official repository adds five local engines, Ollama, Exo, Jan.ai, LM Studio and vLLM, which means a board can run end to end on your own hardware with no prompt ever leaving the machine.
Retrieval sits alongside. Documents and websites can be attached as RAG sources, indexed in a local IndexedDB vector store, with embeddings computed either locally through Transformers.js or through OpenAI. API keys and sensitive values are held in encrypted local storage. The repository also documents custom model support, adjustable inference parameters, per-model metrics such as tokens per second, vision support for images in chats, and multiple boards, alongside interface controls for theme colour, grid snapping, panning and zooming.
Curiso ships in two forms. The v1 desktop application for Windows, macOS and Linux, built with Tauri, is open source under the MIT licence and free to download from GitHub. The web application at app.curiso.ai is the subscription product, with three tiers and a three-day trial that needs no card. Subscriptions are settled in CUR, the project's Solana utility token, and a third of the CUR collected is burned from the supply. The project was created by Carsen Klock, and its domain was registered in January 2025.
What it does
- Hold conversations with large language models on an unlimited canvas instead of a single thread
- Wire nodes together so context flows from one to the next without being retyped
- Switch between twelve cloud providers and five local inference engines using your own API keys
- Attach documents and websites as RAG sources indexed in a local vector store
- Capture ideas in note nodes and feed them straight into a chat
- Run deep research passes on complex subjects
- Publish a board openly or share it with named collaborators
When to use Curiso / When not to
A quick filter to help you decide if Curiso is the right fit.
When to use Curiso
- Advanced LLM users who want to steer a conversation across branches rather than scroll a single linear thread
- Researchers and analysts who put the same question to several models and compare the answers side by side
- Privacy-conscious users who would rather run inference locally through Ollama, Exo, Jan.ai, LM Studio or vLLM than send prompts to a hosted service
- Developers and tinkerers who want auditable MIT-licensed code, custom model endpoints and adjustable inference parameters
- People who already pay for provider APIs and refuse yet another per-model subscription on top
When not to use Curiso
- Anyone expecting a ready-to-use chatbot: you must supply your own API keys or install a local engine first
- Teams with a compliance checklist, since the site never mentions the GDPR, publishes no DPA and names no subprocessors
- Buyers who will not touch cryptocurrency, because subscriptions are settled in the CUR token on Solana
- Users who need a native mobile app, as only a mobile-friendly web app is offered
- Anyone under 18, who is excluded by the terms of service
How to use Curiso
A typical end-to-end flow, from setup to results.
- Decide which route you want: the free open-source desktop app, or the subscription web app
- For the desktop route, download the latest release from the GitHub repository for Windows, macOS or Linux and install it
- For the web route, open app.curiso.ai and start the three-day trial, which asks for no card
- Add your API keys for the cloud providers you intend to use, or point Curiso at a local engine such as Ollama, Exo, Jan.ai, LM Studio or vLLM
- Create a board and place your first chat node, then choose the provider and model it should call
- Add note nodes for raw ideas and connect them to chat nodes so the context carries over
- Wire chat nodes to each other to branch a line of thought without losing what came before
- Attach documents or websites as RAG sources and pick a local or OpenAI embedding model
- Adjust inference parameters and watch the per-model metrics such as tokens per second
- Share the finished board publicly or with named collaborators
Pros & Cons
Pros
- The node canvas genuinely solves context repetition: a branch inherits what came before instead of being retyped
- Twelve cloud providers and five local engines behind one interface, with no vendor lock-in
- Bring-your-own-key pricing means no reseller margin on tokens; you pay providers at their own rates
- Fully local operation is possible, with local inference, local embeddings and a local vector store
- The v1 desktop app is open source under MIT and available on all three desktop operating systems
- API keys are stored encrypted on the machine rather than on a server
- A three-day trial with no card required, and a referral scheme that extends the subscription
Cons
- Subscriptions are payable in the CUR token on Solana, which requires a crypto wallet and exposes you to its price swings
- No legal entity, postal address or identifiable jurisdiction appears anywhere on the site
- No GDPR mention, no DPA, no subprocessor list and no security certification such as SOC 2 or ISO 27001
- The only published address is privacy@curiso.ai; support runs through Discord and Telegram rather than a support desk
- No public API and no documentation for one, so the tool cannot be wired into other systems
- No native mobile app, and no published data retention period
- A young project, with the domain registered in January 2025, and a single publicly identified developer behind it
Pricing & Plans
There is no permanent free plan on the hosted service, but a three-day free trial gives full access to core features and requires no credit card. The lowest paid entry point is the Bronze plan at USD 10 per month, or USD 100 per year, which amounts to ten months paid for twelve. Two higher tiers follow at USD 40 and USD 100 per month. Subscriptions are settled in the CUR token on Solana rather than in fiat currency, so the amount actually paid depends on the token's rate at the time. Model usage is not included: users supply their own provider API keys, or pay nothing at all when running locally. Separately, the v1 desktop application is free and open source under the MIT licence.
- Free 3-Day Trial — full access to core features
- no credit card required
- Bronze Plan — USD 10 per month or USD 100 per year — infinite canvas
- early access to the web app
- standard RAG
- board sharing
- Silver Plan — USD 40 per month or USD 400 per year — all Bronze features
- data backup and recovery
- board sharing
- early access features
- Gold Plan — USD 100 per month or USD 1000 per year — all Silver features
- image
- video and audio nodes
- board sharing
- early access features
Data, GDPR & hosting
A consolidated view of how Curiso handles your data.
GDPR overview
There is no mention of the GDPR anywhere on the site — the regulation is never named, neither to claim compliance nor to deny it. No Article 27 representative is designated, no data protection officer is named, no data processing agreement is offered and no subprocessor list is published. The privacy policy does list rights that read like GDPR rights without ever tying them to a legal basis: access to your personal data, correction or deletion, opting out of non-essential collection, portability, and withdrawal of consent at any time. A single channel is given for these requests, privacy@curiso.ai, with a stated response time of 48 hours. The terms invoke applicable international laws and arbitration in the jurisdiction of incorporation, without naming a country.
Who owns the data?
The terms of service draw a clear line. Content you create stays yours: user-generated content remains user property. Everything belonging to the platform itself is protected by copyright, and you receive only a limited licence to use it, with no right to reproduce or redistribute it. Nothing in the published terms transfers ownership of your boards, notes or conversations to the publisher, and no clause grants a licence over them for promotion or model improvement. On the desktop side the point is largely moot: API keys and sensitive values sit in encrypted local storage, and boards can be kept entirely on your own machine.
Reuse rights
Because your boards, notes and conversations remain your property, you can reuse, publish, quote or commercialise them without asking permission, subject to whatever your model provider's own terms impose on generated output. The platform grants sharing features for exactly this purpose: a board can be published openly or opened to named collaborators. The limit runs the other way. Curiso's own content, branding and interface stay under copyright, you hold only a limited licence to use them, and unauthorised reproduction or distribution is forbidden. Note that CUR token transactions are recorded on a public blockchain and are therefore readable by anyone, permanently and irrespective of any request you make.
Data retention & training
Hosting summary
The site publishes no hosting country and no hosting region, and names no cloud provider for the web application. The domain sits behind Cloudflare, whose address is anycast, so the location returned by geolocation identifies a network node rather than a data centre. The privacy policy offers only the phrase secure blockchain infrastructure, which says nothing about where account data lives. Two elements are known indirectly: PostHog analytics is called on a United States endpoint, and Sentry error tracking is present in the application bundle, neither of them declared in the policy. Against that, a substantial part of the data never reaches a server at all. The vector store is a local IndexedDB, embeddings can be computed on the machine, and API keys are held in encrypted local storage. Running a board against a local engine such as Ollama or LM Studio keeps prompts on the device entirely. Anyone with a jurisdictional requirement should ask directly before subscribing.
Things to keep in mind
Risks and trade-offs to weigh before adopting Curiso.
- Subscriptions are paid in a volatile crypto token, and the terms state plainly that there are no refunds for token purchases
- The terms disclaim any guarantee of service availability and reserve the right to discontinue the platform at any time
- No liability is accepted for a compromise of your wallet, which shifts a real financial risk onto you
- With no named company, no address and arbitration in an unspecified jurisdiction, you would have little practical recourse in a dispute
- PostHog analytics and Sentry error tracking load in the application without being named in the privacy policy
- A canvas that never fills up invites hoarding: boards can accumulate half-finished branches faster than they produce conclusions
- Branching a question across several models makes it easy to shop for the answer you already wanted, and to mistake agreement between models for corroboration
Setup & Integrations
Technical difficulty
Moderate, and it depends on the route. The web app needs only a subscription and no installation; the desktop app is a normal download and install on Windows, macOS or Linux. The real hurdle for a non-technical user is the same either way: you must obtain API keys from one or more providers and paste them in. Running locally means installing and operating Ollama, Exo, Jan.ai, LM Studio or vLLM yourself, and Windows users hit a documented CORS problem with Ollama. Building from source additionally requires Bun, Rust and the Tauri CLI.
Deployment
Integrations
Behind Curiso
Fundraising
Resources
All the official URLs gathered for verification and reference.
Frequently asked questions
Do I need my own API keys to use Curiso?
Is there a free way to use Curiso?
How much does a subscription cost?
How are subscriptions actually paid?
Which operating systems and devices are supported?
Does Curiso offer an API?
Is my data used to train AI models?
How long is my data kept, and can I have it deleted?
Is there a referral or affiliate scheme?
Is the code open source?
Should you pick Curiso?
Curiso is a tool for a specific kind of user, and it makes no attempt to hide it. The node canvas is a real idea rather than a skin on a chat window: branching a line of thought while carrying context forward is something linear interfaces handle badly, and Curiso handles it natively. Around that sits an unusually open architecture — twelve cloud providers and five local engines under one roof, your own API keys, local embeddings and a local vector store, and a v1 desktop application published under the MIT licence for all three desktop platforms. For someone who already pays for model APIs, or who wants prompts never to leave the machine, that combination is genuinely hard to find elsewhere.
The reservations are just as clear, and they are not about the software. No legal entity, no postal address and no identifiable jurisdiction appear anywhere on the site. The GDPR is never mentioned, no data processing agreement or subprocessor list is published, and no retention period is stated, while analytics and error tracking run in the background without being named in the privacy policy. Subscriptions are settled in a Solana token, which adds price volatility and a wallet requirement to what is otherwise a straightforward software purchase, and the terms disclaim availability, refunds and wallet security in fairly absolute terms.
The sensible reading is to treat the two distributions differently. The free MIT desktop app, driven by your own keys or a local engine, carries little of this risk and is worth trying on its merits. The paid web tiers ask for more trust than the published information currently supports — a fine bet for an individual, a harder one for an organisation with a compliance function.
- Choosing a selection results in a full page refresh.
- Opens in a new window.