Web data and models, running where you choose.
Get useful web data. Run models on your own infrastructure. Web data for the questions and pages your application needs, and model access that runs where you choose.
Hosted APIs bill nothing while in beta, pilot and alpha. Software you run costs nothing from Vitna.

Ranked pages from Vitna's own index, each with its URL, an excerpt and when it was read.
A question in words
Hosted by Vitna, on its own index
Ranked pages, each with its URL, an excerpt, and when Vitna read it
Asks for A key, by invitation
What do you need?
Six sentences, each mapped straight to the product that answers it and where setup starts.
- 01Private beta

I need search results.
Ranked pages from Vitna's own index, each with its URL, an excerpt and when it was read.
- 02Hosted pilot

I need page content or structured fields.
A URL in, Markdown, text or the JSON fields you name out, from one page or a whole site.
- 03Alpha

I need browser-rendered content.
A fresh browser runs the page's JavaScript and returns the HTML, text and links it rendered.
- 04On npm

I need access to cloud models.
One OpenAI-compatible endpoint on your machine that serves exactly the model you name, on your provider keys.
- 05On npm

I need controlled model routing.
The same endpoint with policies, fallbacks and a record of every request, applying a decision you accepted.
- 06Early release

I need local inference.
An engine that streams an open model off your own drive onto your CPU or NVIDIA GPU.
What goes in, what comes out, and where it runs.
Every product laid out as a single request pipe so you can compare hosted web APIs, cloud gateways and local inference side by side.
Find and retrieve web data
Which tool fits the input you have and the output you need.
- Takes
A question in words
Web search APISearchPrivate betaHosted by Vitna, on its own indexReturnsRanked pages, each with its URL, an excerpt, and when Vitna read it
Asks for A key, by invitationRequest a Search key - Takes
A URL, a site, or a JSON Schema
Page reading and extraction APIsWeb APIsHosted pilotHosted by VitnaReturnsMarkdown, text, or the fields you named; crawls, maps and change alerts
Asks for A pilot key, by invitation while signup is closedRequest a Web API key - Takes
A URL whose content needs a browser
Browser rendering APIUnblockerAlphaHosted by Vitna, a fresh browser per requestReturnsThe HTML, text and links the page rendered, with a receipt of each attempt
Asks for An alpha key, on requestRequest alpha access
Access hosted models
Direct access on your own keys, or policies, fallbacks and controls on top.
- Takes
An OpenAI- or Anthropic-style request
Model gateway on your own keysRelayOn npmOn your machine, at 127.0.0.1:8787ReturnsThe provider's answer, and a ledger line on your disk
Asks for Node 22.13 or later and a provider keyInstall Relay - Takes
The same request, from your services
Self-run model gateway with policiesRuntimeOn npmOn your servers or your cloud accountReturnsThe answer from the lane your signed policy chose, with fallbacks and a record of every call
Asks for The install, provider keys and a signed policyInstall Runtime
Run open models locally
Supported models and hardware, the install, and what has been measured.
- Takes
An OpenAI-style request
Local inference engineAnchorEarly releaseOn your machine, at 127.0.0.1:8765ReturnsAn answer from an open model on your own CPU or NVIDIA GPU
Asks for Windows or Linux on x86-64 through the installer; a binary for Apple siliconGet Anchor
Nine layers. Four of them live today.
From the models at the bottom to the apps at the top. A status changes only when the software does.
- 4 live
- Somebody outside can use it today.
- 4 partly live
- Part of the layer runs, and the rest does not.
- 1 in development
- The code exists. Nobody outside can use it yet.
- 0 planned
- Nothing runs.
- 09
AppsPartly live
Chat, research and agents, as software a person opens.
Today The Vitna app, an alpha on the person's own OpenRouter key: chat, research, images, and a Code mode that reads a folder and proposes edits. No agents yet.
- 08
SurfacesLive
Where people and software meet the stack.
Today vitna.ai, the console, a public API that needs no key, the runtime's command line, and keyed APIs for Search, the Web APIs and the Unblocker, with an MCP endpoint.
- 07
Accounts and keysPartly live
Sign in, open an organisation, add people, mint keys, pay.
Today Sign-in and organisations. Each product keys its own API: the console's push keys and members are still an operator's command, the Web APIs open accounts of their own, and checkout is not switched on.
- 06
DecisionsLive
Decide when a production stack should change, show the arithmetic, and let a person rule.
Today The record of prices, read from providers' own pages with its days signed, and the engine that turns a watched move into a brief.
- 05
EvidencePartly live
Every step from a published price to a finished run, signed, and pointing at the one before.
Today Signed days, signed policies and the runtime's ledger, each signed on its own.
- 04
Spend and permissionIn development
An agent spends and acts only inside authority somebody delegated to it.
Today Built as its own protocol and ledger, and not yet wired into the stack or audited.
- 03
Agents and toolsPartly live
Multi-step work with tools, in a sandbox, with approvals.
Today Agents read the web through Search, the Web APIs and their MCP endpoint, and any agent that speaks the OpenAI dialect can work through a runtime lane. Code reads a folder and proposes edits in the app's alpha, and runs no commands.
- 02
The gatewayLive
One OpenAI-compatible endpoint on your side, serving the decision per request.
Today vitna-runtime on npm: signed lanes, ordered fallbacks, staged rules and a ledger, and no call to vitna.ai.
- 01
Reach models on your own keys, in the cloud or on your own hardware.
Today OpenAI, Anthropic and everything OpenRouter lists, on your keys. Anchor runs open models on your own CPU or NVIDIA GPU, from release v0.2.1.
