Dim0
Dim0 is an open-source, real-time collaborative AI canvas where rich notes, sketches, code, documents and generated React mini-apps share one infinite board, and a board-aware agent reads that board before it searches, runs code and writes results back.
What is Dim0?
Dim0 is a thinking canvas: a single infinite board on which rich notes, freehand shapes, executable code nodes, uploaded documents, charts, nested boards, presentation frames and AI-generated mini-apps all live side by side. The differentiator is where the AI sits. Instead of a chat sidebar bolted onto a document, the agent is a resident of the board. It reads the whole graph and whatever you have selected, takes multi-step actions, then writes its results back as editable nodes exactly where you are working.
Those actions are concrete. The agent searches the web and fetches pages, runs code in sandboxes, queries documents you have uploaded through semantic retrieval, drafts and revises notes in place, and turns loose notes into mind maps, schemas or diagrams. It also generates mini-apps: real interactive React applications, a calculator, a chart, a visualiser or a quiz, which persist on the board, read the data sitting next to them, can be used live by your team, and can be opened, edited and exported as code.
Every board is multiplayer. Live cursors, presence, shared agents and concurrent editing behave the same whether you work alone or fifty people deep, and conflicts are resolved with operational transforms and last-write-wins rather than CRDT, an approach the publisher attributes to Figma and Excalidraw. Underneath sits canvas-harness, a separately released open-source engine that paints to a canvas instead of the DOM: the publisher measures ten thousand visible nodes panning at roughly eighty frames per second on an M1 laptop.
The product is deliberately unlocked. The code is MIT-licensed and self-hostable with a local Postgres database, a local vector database and your own model keys, while notes are stored as plain Markdown and boards export cleanly. Eight models are interchangeable at any moment, Claude, GPT, Gemini, Mistral, DeepSeek, Qwen, Kimi and GLM, alongside eight light and dark themes and a whiteboard layer with shapes, arrows, media and a very large icon library.
Two caveats frame all of this. The product is young, announced as early access at version 0.3, with a free tier the publisher openly calls constrained. And its legal footing is thin: no company entity, no postal address and no mention of European data protection rules anywhere on the site.
What it does
- Read the whole board, including the nodes you have selected, before taking any action
- Generate interactive React mini-apps such as calculators, charts, visualisers and quizzes as live nodes
- Search the web and fetch pages during a multi-step task
- Run code inside sandboxes and bring the results back onto the canvas
- Draft and revise rich notes in place, with maths, code and sub-pages
- Turn loose notes into mind maps, schemas or diagrams in a single gesture
- Query uploaded documents through semantic retrieval and recall stored context
When to use Dim0 / When not to
A quick filter to help you decide if Dim0 is the right fit.
When to use Dim0
- Product and design teams who map an idea together and want live cursors, presence and a shared agent on the same board
- Developers and technical teams who need sandboxed code nodes, exportable React mini-apps and a Docker self-hosting path
- Privacy-conscious users and organisations with internal constraints, who can inspect the MIT code, run it themselves and bring their own model keys
- Researchers, doctoral candidates and students who want web search, uploaded documents queried through retrieval, and notes that stay portable Markdown
- Consultants and freelancers who go from a spatial rough draft to a presentation without changing tool, using frames and presentation mode
When not to use Dim0
- Teams whose work is fundamentally relational databases, rollups and filtered views, which the publisher states Dim0 has nothing comparable to
- Large organisations that require SSO, governance and compliance tooling, explicitly conceded to Miro in the vendor's own comparison
- People who need a native mobile experience, since the site describes a browser product where tablets work well and phones are mainly read-friendly
- Users who want local-first, offline, single-player note keeping, a priority the vendor openly says Dim0 does not try to serve
- Buyers who need a mature platform with a template library, workshop facilitation tools and business integrations such as Jira or Slack
How to use Dim0
A typical end-to-end flow, from setup to results.
- Open app.dim0.net and sign in; no payment card is requested to start
- Create a board, then drop nodes onto it: notes, shapes, code, documents, mini-apps or nested boards
- Select the context you want the agent to work from before you prompt it
- Type your prompt in the board's input, enabling the web search step when the task needs it
- Let the agent answer by placing editable nodes on the canvas: a note, a mind map, a chart or a mini-app
- Invite collaborators; the board becomes multiplayer with live cursors and presence
- Group work into nested boards to keep a large project organised
- Drop frames around the parts you want to show, then run presentation mode straight from the board
- Download notes as Markdown or export the board when you want to take the work elsewhere
- To self-host instead, clone the repository, copy .env.sample to .env, set the three required API keys, then run make pull and make run and open localhost:3000
Pros & Cons
Pros
- Notes, canvas and AI on one surface, with nothing to copy and paste between tabs
- The agent writes editable nodes onto the board instead of burying answers in a chat thread
- MIT-licensed and self-hostable, so the code, the infrastructure and the model keys can be yours
- No proprietary format: notes are plain Markdown and boards export cleanly
- Explicit privacy commitments, with no training on content, no data sale and no behavioural telemetry
- Eight models switchable at will, and AI that is not metered by credits on the paid tiers
- Low entry price at EUR 6.99 per month, with real-time collaboration included at every tier
Cons
- Very young product: domain registered in February 2026, first archived capture in May 2026, version 0.3 in early access
- No legal entity published anywhere: no company name, no postal address, no registration number
- No mention of GDPR, no data processing agreement, no subprocessor list and no declared hosting country
- No security certification and no trust page; no customer-held keys or end-to-end encryption in the cloud
- No relational databases, no enterprise governance such as SSO, and no workshop facilitation tooling
- Browser-only in practice for mobile use, with phones described as mainly read-friendly
- Free tier openly described as limited and subject to change, with top-tier models reserved for the Plus plan
Pricing & Plans
A permanent free plan is available at no cost, and the vendor states that no payment card is required to begin. The cheapest paid tier is Basic, priced at EUR 6.99 per month, with Plus at EUR 11.99 per month. Self-hosting the MIT-licensed code carries no licence fee, infrastructure and model keys remaining at the user's own expense.
- 50 AI requests per day and 750 per month
- 5 boards
- up to 5 collaborators per board
- 3 documents and 10 mini-apps per board
- Lite models only
- community support
- 150 AI requests per day and 3
- 000 per month
- unlimited boards
- up to 10 collaborators per board
- 10 documents and 20 mini-apps per board
- Lite models only
- standard support
- unlimited AI requests
- unlimited boards
- up to 20 collaborators per board
- 25 documents and 100 mini-apps per board
- frontier models including GPT
- Claude and Gemini
- priority support
- full source on GitHub
- your own infrastructure and data
- bring your own model keys
- unlimited collaborators
- no caps
Data, GDPR & hosting
A consolidated view of how Dim0 handles your data.
GDPR overview
There is no GDPR mention anywhere on this site. A search across every collected page, privacy policy included, returns no occurrence of GDPR, of data protection regulation, or of any equivalent framework. No legal basis is stated, no controller is named, no data protection officer or Article 27 representative is designated, no data processing agreement is offered, no subprocessor list is published and no hosting country is declared. Data subject rights are not enumerated: the policy says only that you may request deletion at any time and points to a single generic address. This silence is notable because the footer places the publisher in Paris, France, so European rules would apply. The substance is nonetheless protective, with no training on content, no data sale and no behavioural telemetry, but the formal apparatus is entirely absent.
Who owns the data?
You keep ownership of the content you create, upload or connect to Dim0, and you are responsible for holding the rights to whatever you bring in and for how you use generated outputs. The publisher commits never to train models on boards, notes, prompts or uploads, neither its own nor anyone else's, never to sell data, share it for advertising or profile you. Prompts and board context reach only the model provider you selected, for the request you sent, and those providers process that data under their own terms; no copies of prompts are kept for training. One reservation matters: the terms name no legal entity, so the controller behind these commitments is not identified.
Reuse rights
Nothing on the site requires permission to reuse your own material. Notes are stored as plain Markdown and can be downloaded and pasted into any other editor, boards export cleanly, and the publisher claims no proprietary format and no lock-in. Generated mini-apps are real React applications whose code you can open, edit and export, and the underlying canvas engine exposes selection exports as PNG and scene exports as Markdown. The code itself is MIT-licensed, so you may fork it, self-host it and plug in your own model keys. The single reservation is editorial rather than legal: the terms state that AI outputs can be wrong, incomplete or inappropriate, and place the duty of review on you before you rely on them or treat them as factual.
Data retention & training
Hosting summary
No hosting country, region or infrastructure provider is declared for user data anywhere on the site, and no jurisdiction is stated in the privacy policy or the terms; the only geographic clue is the footer line placing the publisher in Paris, France. What the policy does describe is the protection applied rather than the location: everything travels over TLS, and boards, notes, uploads and database records sit on storage that encrypts data at rest by default. The publisher explicitly does not offer customer-held keys or zero-knowledge end-to-end encryption and points users who need either towards self-hosting. Content relevant to a request is sent to the model provider you selected, among Anthropic, OpenAI, Google, Mistral, DeepSeek, Qwen, Moonshot and Z.ai, which process it under their own terms. In the self-hosted mode, everything stays on your infrastructure with a local Postgres database and a local vector database, and privacy, retention and encryption then depend entirely on how you deploy the system.
Things to keep in mind
Risks and trade-offs to weigh before adopting Dim0.
- The publisher is not legally identified: no company name, address or registration means no named counterparty if something goes wrong
- Board content is forwarded to third-party model providers who apply their own terms, outside the publisher's control
- No customer-held keys and no end-to-end encryption in the cloud, a limit the vendor acknowledges while pointing to self-hosting
- AI outputs can be wrong, incomplete or inappropriate, and the terms place the entire burden of review on the user
- The service is supplied as-is and as-available during early access, which makes it a risky dependency for critical work
- Free-tier quotas are openly described as subject to change while the publisher runs on a small budget
- A canvas that thinks for you can quietly erode your own habit of structuring ideas if the agent always drafts and organises first
Setup & Integrations
Technical difficulty
Effectively none in the cloud: create an account, no installation, no payment card, and start dropping nodes on a board. The only learning curve is the spatial interface itself, since there are no guided templates. Self-hosting is a developer task: Docker and Docker Compose, a repository clone, an .env file with three paid API keys, then two make commands. Running from source additionally needs Node.js LTS and the uv Python tool. The desktop build installs from a one-line script or a release installer, though the binary is only ad-hoc signed.
Deployment
Integrations
Behind Dim0
Resources
All the official URLs gathered for verification and reference.
Alternatives
Tools that compete with or complement Dim0.
Frequently asked questions
What exactly is Dim0?
Is Dim0 open source, and can I run it myself?
How much does it cost?
Which AI models can I use?
What can the agent actually do?
Is my content used to train AI models?
Does real-time collaboration work, and how?
Is there a mobile or desktop application?
Can I get my work out of Dim0?
Is Dim0 GDPR compliant?
Should you pick Dim0?
Dim0 makes an unusually clear promise and, on the evidence of its own pages, keeps most of it. The canvas is the interface rather than a feature bolted onto a document, the agent lives on the board and writes editable nodes back onto it, and the mini-apps are genuine React applications rather than snippets trapped in a chat thread. The engineering behind that claim is unusually visible: a separately released rendering engine with measured performance, a documented operational-transform sync model, and a public repository with a full self-hosting path. The anti-lock-in stance is consistent throughout, from the MIT licence to Markdown notes and bring-your-own model keys, and the pricing is legible and low, with three tiers between zero and twelve euros and no credit meter on AI usage.
The reservations are equally clear. This is a very young product: the domain dates from February 2026, the first archived capture from May 2026, and the vendor itself labels the phase early access at version 0.3 with a free tier it calls constrained. The privacy commitments are strong in substance but bare in form, with no legal entity named, no postal address, no certification and no mention at all of European data protection rules on a site whose footer places the publisher in Paris. The site and the official repository also disagree on whether a desktop application exists. Anyone weighing Dim0 for team or regulated work should treat those gaps as open questions rather than oversights.
For individuals and small teams who think spatially, who want AI acting on real content in context, and who value the ability to walk away with their data, Dim0 is worth a serious trial at no cost. For organisations that need governance, contractual guarantees and maturity, it is a project to watch rather than adopt today.
- Choosing a selection results in a full page refresh.
- Opens in a new window.