How much does Dyad cost?

Dyad Free costs nothing with the user's own API key and includes 20 Basic Agent messages a day. Pro is $20 a month for 200 AI credits with Pro modes, the Pro agent and Academy access, and Max is $79 a month for 900 credits. Credits reload at the same price, unused credits roll over up to 3,600, and an orchestration fee of 1.5 credits per million tokens applies. No annual, team or enterprise plan is listed.

What are the best alternatives to Dyad?

Based on documented capability coverage across the Agentic Index 14 point taxonomy, the closest alternatives to Dyad among coding agents are Macaly, Caffeine, Cosine, Goose. Each is graded on the same public evidence, and no vendor pays for placement.

Can Dyad be self hosted or run in a private VPC?

Full under the Agentic Index verification standard. Dyad publishes public evidence of self hosted or private VPC deployment, along with data residency options for controlling where agent data is processed and stored. Evidence basis: Dyad runs entirely on the user's computer, building and running apps locally with an optional Docker runtime, and with Ollama or LM Studio models nothing needs to leave the machine. Built apps live in local Git repositories and publish to h. Graded from public evidence only against the Agentic Index 14 point capability taxonomy. No vendor pays for placement.

Does Dyad require human approval before an agent acts?

Full under the Agentic Index verification standard. Dyad publishes public evidence of human approval steps and guardrails, so an agent action can be held for review before it executes. Evidence basis: Every change Dyad makes is a git commit that can be diffed side by side, undone, retried or restored, with restores adding a new version rather than erasing history. Each tool runs on an always, ask or never setting with accept once, accept. Graded from public evidence only against the Agentic Index 14 point capability taxonomy. No vendor pays for placement.

Can you bring your own model or route between providers with Dyad?

Full under the Agentic Index verification standard. Dyad publishes public evidence of bring your own model and bring your own key (BYOK), and of routing tasks between models or providers rather than being locked to one. Evidence basis: Nine providers are built in, OpenAI, Anthropic, Google Gemini, OpenRouter, Vertex AI, Azure OpenAI, xAI, Bedrock and MiniMax, along with any OpenAI compatible endpoint, local models through Ollama or LM Studio, and ChatGPT or Claude Code su. Graded from public evidence only against the Agentic Index 14 point capability taxonomy. No vendor pays for placement.

Back to vendors

Dyad

Also known as: Dyad.sh, Dyad Pro, Dyad Tech

Visit site

Coding agentindependentVerified 2026-11-08

Entry priceFree with the user's own API key, or Pro at $20 a month for 200 AI credits.Full pricing detail

Dyad is a desktop app builder from Dyad Tech, Inc. that turns a chat description into a working JavaScript web app on the user's own machine, with a live preview, Supabase or Neon for the database and auth, and one click publishing to Vercel, Cloudflare and other hosts. Its core is open source under Apache 2.0, it runs on any provider's models or local ones through Ollama, and it is free with the user's own keys; Pro costs $20 a month.

Dyad is made by Dyad Tech, Inc., a US company whose creator is Will Chen, and is distributed as a desktop app for macOS and Windows, with Linux builds marked experimental. The code outside its Pro directory is open source under Apache 2.0, while the Pro agent, editing and visual editor code uses the Functional Source License, which bars competing commercial use and turns into Apache 2.0 two years after release. The GitHub repository has about 22,000 stars, and every app Dyad builds is a regular Git repository in a folder on the user's computer.

A builder describes an app in chat and Dyad writes it, by default as a React and Vite app or from a Next.js, community or experimental Portal template, then runs it locally with a live preview. Agent mode plans with a task list, writes code, checks the preview, reads server and browser logs and fixes errors over up to 100 tool steps, with a code explorer and an implementer subagent, and several agent chats can change one app at once.

It connects Supabase or Neon by OAuth to create projects, run SQL and manage auth, syncs and branches on GitHub, publishes to Vercel, Cloudflare Workers, GitLab or Coolify, and calls any MCP server over stdio or HTTP. Every change is a version that can be diffed and restored, tools run on always, ask or never consent, and schema changing SQL asks first. A TypeScript check fixes errors automatically, and experimental testing and security review panels run and fix tests and findings.

Dyad works with OpenAI, Anthropic, Google, OpenRouter, Vertex AI, Azure OpenAI, xAI, Bedrock, MiniMax and any OpenAI compatible provider, local models through Ollama or LM Studio, and ChatGPT or Claude Code subscriptions. With the user's own keys, prompts and code never pass through Dyad's servers; Pro traffic is proxied through them, Dyad does not train on user data, and telemetry is opt in. It names no security certification, single sign on or team roles.

Dyad Free costs nothing with the user's own API key and includes 20 Basic Agent messages a day, Pro costs $20 a month for 200 AI credits with Smart Context, the Pro agent, web search and website cloning, and Max costs $79 a month for 900 credits, with unused credits rolling over up to 3,600 and an orchestration fee of 1.5 credits per million tokens. It fits builders who want to own their code and run everything on their own machine and models, and developers who want an open source alternative to hosted app builders. It fits less well for teams that need hosted collaboration, enterprise security, or apps outside the JavaScript stack.

Vendor details

Canonical URL

https://www.dyad.sh

Subcategory

Open source desktop app builder that builds JavaScript web apps locally on any model, with Supabase or Neon backends and publishing to the customer's own hosts

Funding status

No funding is stated on Dyad's own pages; Dyad Tech, Inc. is named as the company behind the product, with Will Chen as its creator.

Company status

independent

Use cases & customers

Primary use cases

building full stack web apps from chat on a local machineconnecting Supabase or Neon for databases and authpublishing apps to Vercel, Cloudflare Workers or self hosted serversbuilding with local models through Ollama or LM Studioreviewing and restoring every AI change as a git version

Target customers

builders who want to own their code and keep it on their machinedevelopers who want an open source alternative to hosted app builderspeople who want to build with local models or their own API keysmakers of full stack web apps on Supabase or Neon

Deployment options

Desktop app on macOS and Windows, with experimental Linux buildsApps run locally, optionally in DockerPublishing to Vercel, Cloudflare Workers, GitLab or CoolifyFully local use with Ollama or LM Studio models

Integrations

Supabase and Neon by OAuth for databases, auth, SQL and project creation; GitHub sync, branches, collaborators, issues and import; publishing to Vercel, Cloudflare Workers, GitLab including self hosted, and Coolify; an MCP client over stdio and HTTP with OAuth and a Hub catalog of servers such as Context7, Playwright and Netlify; and ChatGPT and Claude Code subscriptions as model sources.

In practice

A freelancer builds a client booking site in Dyad on her laptop with her own Anthropic key. Agent mode plans the pages, connects a new Supabase project for auth and bookings, asks before running a migration that changes the schema, and she publishes the result to her client's Vercel account.

A developer working offline runs Dyad against a coding model in LM Studio, so prompts and code stay on his machine, and syncs the finished app to a private GitHub repository for his team to review.

A small agency on Dyad Pro imports an existing Next.js repository, uses Smart Context to work on the large codebase, runs the experimental security review and fixes its findings, then deploys to its own Coolify server over SSH.

Sources & related URLs

Related / legacy domains

https://www.dyad.sh/docs https://www.dyad.sh/pro https://github.com/dyad-sh/dyad https://www.dyad.sh/hub

Research sources

Agentic Index coverage score

10.5 / 14 capabilities · 75%

Integrations & Tool Calling	Full
Supabase and Neon connect by OAuth, and the agent creates Supabase projects, runs SQL, reads table schemas and sets up auth, while Neon brings development and production branches, schema migration to production and environment variables synced to Vercel. GitHub sync covers branches, collaborators, conflict resolution and issues through a plugin, publishing reaches Vercel, Cloudflare Workers, GitLab and Coolify, and any MCP server can be added over stdio or HTTP with OAuth. Sourcedyad.sh/docs/integrations/supabaseread 2026-11-08
Workflow Orchestration	Full
Agent mode keeps a task list and works through up to 100 tool calls, writing code, checking the preview, reading logs and fixing errors until the task is done, after a plan mode or app blueprint when wanted. A code explorer and an implementer subagent split the work, an experimental Auto Sidekick lets a larger model supervise a smaller one, and several agent chats can change the same app at once while queued tasks pause and resume. Sourcedyad.sh/blog/ai-agent-mode-explainedread 2026-11-08
Knowledge Grounding & RAG	Partial
On Pro, Smart Context uses a light model to pick the most relevant files and Deep Smart Context searches the codebase several times per answer, while a code explorer subagent and grep and code search tools find code on demand. Pro agents also search the web and crawl pages, and library documentation can come in through the Context7 MCP server. No index of the codebase or document library persists between sessions. Sourcedyad.sh/docs/guides/ai-models/pro-modesread 2026-11-08
Human Oversight & Guardrails	Full
Every change Dyad makes is a git commit that can be diffed side by side, undone, retried or restored, with restores adding a new version rather than erasing history. Each tool runs on an always, ask or never setting with accept once, accept always or decline prompts, MCP calls need approval by default, and SQL that changes the schema or deletes data asks first, though Pro users can let an AI reviewer approve safe MCP calls. Sourcedyad.sh/docs/guides/versioningread 2026-11-08
Security, Identity & Governance	Partial
With the user's own API keys, prompts and code go straight to the provider and never through Dyad's servers, while Pro traffic is proxied through Dyad and may be logged for abuse or outages. Dyad does not train on user data, telemetry is opt in and excludes prompts and code, connection tokens stay on the user's computer, releases are signed, and an optional Docker runtime and a package firewall add isolation, but there is no certification, single sign on or team role. Sourcedyad.sh/docs/policies/privacy-policyread 2026-11-08
Observability & Auditability	Partial
A system messages drawer, the preview's console and runtime logs that survive restarts, and a token usage panel show what happened in a session, along with test results, deployment status and subagent activity in chat. Usage appears in the Academy portal, and a bug report bundles logs into a GitHub issue, but there is no audit log, exportable run history or tracing across sessions. Sourcedyad.sh/docs/guides/debuggingread 2026-11-08
Memory & State Persistence	Partial
Pro agents can search earlier chats to bring back past work, long chats compact or summarize into a new chat, and favorite chats and versions carry notes. An AI_RULES file in the project joins the system prompt and a prompt library stores reusable prompts, but these are instructions, and no memory of a builder's preferences or corrections builds up or can be viewed and deleted. Sourcedyad.sh/docs/guides/ai-rulesread 2026-11-08
Deployment & Data Residency	Full
Dyad runs entirely on the user's computer, building and running apps locally with an optional Docker runtime, and with Ollama or LM Studio models nothing needs to leave the machine. Built apps live in local Git repositories and publish to hosts the customer chooses, including Vercel, Cloudflare Workers, GitLab and a self hosted Coolify server, or build for iOS and Android through Xcode and Android Studio, and Dyad hosts no apps of its own. Sourcedyad.sh/docs/faqread 2026-11-08
Prebuilt Agents / Templates / Packs	Full
New apps start from a React and Vite template, a Next.js template with shadcn and Tailwind, an experimental Portal template with Neon and Payload CMS, or community templates with a set process to submit more. A Hub catalog lists ready MCP servers such as Context7, Playwright, Auth0 and Netlify, and a theme generator, a prompt library and inspiration prompts give further starting points. Sourcedyad.sh/docs/templatesread 2026-11-08
Triggers & Channel Coverage	Partial
Every run starts from a person in the desktop app, by typing or speaking in chat with images, PDFs or zip files attached, or by pressing Fix with AI, Run Security Review or Fix Issue, and deep links can add Hub items or a Pro key. No schedule, event, webhook, command line or public API starts an agent, so nothing happens unless someone has the app open and asks. Sourcedyad.sh/docs/guides/chattingread 2026-11-08
Model Flexibility & Routing	Full
Nine providers are built in, OpenAI, Anthropic, Google Gemini, OpenRouter, Vertex AI, Azure OpenAI, xAI, Bedrock and MiniMax, along with any OpenAI compatible endpoint, local models through Ollama or LM Studio, and ChatGPT or Claude Code subscriptions. Auto picks a keyed provider, Pro adds routing modes such as Super Value, a balanced Auto and Turbo Edits with a second editing model, and thinking effort can be set per model. Sourcedyad.sh/docs/guides/ai-modelsread 2026-11-08
APIs / SDKs / MCP Extensibility	Partial
The code outside the Pro directory is open source under Apache 2.0, so a team can fork and change the builder, and the Pro agent code uses the Functional Source License, which turns into Apache 2.0 after two years. Dyad offers no public API, SDK, command line tool or MCP server of its own for driving the builder, and its MCP client and plugins reach outward to other services rather than letting another program drive Dyad. Sourceraw.githubusercontent.com/dyad-sh/dyad/main/licenseread 2026-11-08
Testing, Debugging & Optimization	Full
Proposed edits run against the TypeScript compiler and errors are fixed automatically up to twice, a problems pane and type check tool catch the rest, and production builds are verified before publishing. An experimental testing panel lets agents run and fix tests, tests can be recorded from the preview and run in parallel in isolated workspaces, and an experimental security review ranks findings by severity with a Fix Issue button. Sourcedyad.sh/docs/guides/security-reviewread 2026-11-08
Browser / Computer-use	Partial
The agent checks the running preview and reads its browser logs, Pro website cloning sends a crawler to a page and takes a screenshot, and screenshots are kept for each app version. Browser control through Playwright, Chrome DevTools or Browserbase comes only from third party MCP servers a user adds, and no Dyad agent is described clicking through an app or driving a desktop. Sourcedyad.sh/blog/ai-agent-mode-explainedread 2026-11-08

The Agentic Index coverage score grades every vendor Full, Partial or Not documented against the same 14 buyer facing capabilities, from public evidence only. Each capability links to how all vendors in the index score on it. How this evidence is graded

Pricing

Free with the user's own API key, or Pro at $20 a month for 200 AI credits.

Monthly subscriptions with AI credits that match model costs, reloads at the same price and an orchestration fee per million tokens.

Free tier

Included quota

Pro includes 200 AI credits a month and Max 900, with unused credits rolling over up to 3,600 that do not expire.

What is public

Free, Pro and Max prices, monthly credits, rollover limits and the orchestration fee.

Billing mechanics

Monthly subscriptions with AI credits matched to model costs, reloads at the same price, rollover up to 3,600 credits and a per token orchestration fee.

Cost watchouts

On the Free plan, model costs on the user's own keys are billed by the provider, and Pro and Max add an orchestration fee per million tokens on top of model costs.

Variable cost rationale

Credits match the model cost of each message, so heavier agent runs on larger models use them faster and need reloads, though Free users pay their model provider directly with their own keys and Dyad adds nothing.

Additional watchouts

No annual, team or enterprise plan, and Pro routes prompts and code through Dyad's servers.

Overage / add-ons

Credits reload at any time at the same price, and Dyad charges 1.5 credits per million tokens as an orchestration fee, or less with a ChatGPT subscription on smaller models.

Sales call required

No, self serve available

Free / trial

Dyad Free costs nothing with the user's own API key, needs no sign up, and includes 20 Basic Agent messages a day with unlimited Build mode.

Lowest paid plan

Pro at $20 a month for 200 AI credits

Commercial notes

Dyad Tech, Inc. gives away the desktop app and sells Pro and Max subscriptions for its proxied models and Pro agent.

Key ambiguities

The Pro page says bring your own key is coming soon while release notes say Pro supports it, and office hours are listed for all Pro users in one place and as a Max perk in another.

Cancellation / refund

Monthly subscriptions cancelled through the Academy portal; refunds are case by case and none once credits are used.

Support SLA / resale

Pro adds Pro modes, the Pro agent, Academy access and early features, and Max adds more credits and prioritized office hours; there is no team or enterprise plan.

Missing data

Annual pricing, team plans and the credit cost of each model.

Agentic Index verified 2026-11-08

Official pricing page Docs

Related vendors

Alternatives to Dyad

The closest documented capability profiles to Dyad among coding agents tracked by Agentic Index, ordered by similarity on the same 14 point evidence the rankings use. No vendor pays for placement.

Macaly10.5 / 14Fuller documented coverage on APIs, SDKs & MCP Extensibility
Caffeine11.0 / 14Fuller documented coverage on APIs, SDKs & MCP Extensibility and Browser & Computer Use
Cosine10.0 / 14Fuller documented coverage on Browser & Computer Use
Goose12.0 / 14Fuller documented coverage on Observability & Auditability and Triggers & Channel Coverage
Zed10.0 / 14Fuller documented coverage on APIs, SDKs & MCP Extensibility
Atoms9.5 / 14Fuller documented coverage on Browser & Computer Use

Similarity is computed from each vendor's Agentic Index coverage score evidence, axis by axis, not from the totals. How this evidence is graded

Compare Dyad with other vendors →Compare all coding agents →See where Dyad ranks →AI agents for software development →

Data confidence: medium