Anthropic

A local Claude API. It serves Messages (api.anthropic.com/v1/messages, streamed and not), token counting and the model list, the calls Dub, Twenty, LibreChat and Claude Code make (journeys/demand.json). Around it are the Console's pages (platform.claude.com) where the API keys are made. The twin runs no model: a turn is the World's scenario's, else a labeled deterministic stub.

Use with an existing app

In an app that already uses this vendor, install this exact release and the product CLI:

npm install --save-dev --save-exact @volter/world@3.0.68 @volter/twin-anthropic@1.0.3
npx volter world init --name my-app --twins anthropic --source anthropic=@volter/twin-anthropic

Review the detected vendor and generated bindings before booting. Read the credential names and limitations below; the World supplies throwaway credentials. Then run your app's own command through the World:

npx volter world up
npx volter world run -- npm test
npx volter world log
npx volter world down

Here npm test is your app's existing command; replace it with your app or test command. down retains state. A later up resumes it; do not reset or initialize again merely to return.

Publisher and catalog contribution instructions: the public publisher guide.

A Protocol 3 pack (publisher guide). Its surface is generated from the OpenAPI document Anthropic's TypeScript SDK is generated from (spec/). Handlers are in src/semantics/messages.ts and src/semantics/models.ts, the key front in src/semantics/around.ts, the scenario adapter in src/semantics/scenario.ts, and pages in src/screens.

world-anthropic serve [--port N] [--root DIR] [--read-only]

What it models

• Keys:
  • Authorization: Bearer or the legacy x-api-key. A missing, mistyped, disabled, deleted or expired key is 401 authentication_error.
  • anthropic-version is required.
  • Every answer carries a request id.
• Messages:
  • the request checked as the API checks it (the model in the workspace catalog and released on the World clock, max_tokens, the thinking budget's rules, manual thinking refused on Claude 4.7 and later, tool_choice naming an offered tool);
  • the answer (msg_…, content blocks, stop_reason, usage) or its server-sent events: message_start, each block's start, deltas (text_delta, thinking_delta and signature_delta, input_json_delta) and stop, message_delta and message_stop;
  • each also as its ?beta=true operation.
• The turn: the scenario's respond (text, toolUses, serverTools, thinking, stopReason), or a stub that:
  • calls the chosen (else the first) tool while no tool result is pending;
  • answers a value the structured output's schema admits (output_config.format, or the beta's output_format);
  • otherwise echoes the last message, labeled.
• Tokens: text at four characters a token, an image at ⌈width/28⌉ × ⌈height/28⌉ visual tokens (read from a PNG's or JPEG's header, at most the model's limit). Token counting (?beta=true, as Claude Code sends it) answers the same count with no message made.
• The prompt cache, per workspace:
  • a prefix per cache_control breakpoint, in the order tools, system, messages;
  • written with its lifetime (5 minutes, or 1 hour) and read while live, each read refreshing it;
  • a prefix under the model's minimum is not cached.
  • usage reports cache_creation_input_tokens, cache_read_input_tokens and cache_creation.
• Models: GET /v1/models, the models an organization can use, newest first, paged by after_id/before_id, each released by the World clock, with its limits, thinking and effort as its documentation page states them.
• The Console: sign-in by an emailed code; the API keys page (create with an expiration, shown once; disable, re-enable, delete).

A scenario's toolUses is an array of { name, input } calls; thinking is a string. For synthetic web search, serverTools is an array of { name: "web_search", input: { query: "..." }, results: [...] }. Each result has type: "web_search_result", url, title, encrypted_content and optional page_age. These scripts describe local fixture output and must match a tool the request offers; tool choice and parallel-use restrictions still apply.

Doors

• POST /_twin/users/{email} {organization}: a person's sign-up, admin of a new organization and its Default workspace.
• GET /_twin/mailbox/{email}: the mail the Console sent them.
• POST /_twin/api-keys {email, organization?, name?}: issue a workspace key without driving the incidental Console.
• POST /_twin/app-credentials {}: issue or resume the World application's key.

Not yet

Message Batches, Files, Skills, Managed Agents, the Admin API, the non-beta token count and the other beta surfaces: no application calls them. Web search can be scripted as synthetic server_tool_use/web_search_tool_result blocks using respond.serverTools; no hosted search runs. Unconfigured hosted tools, code execution and trusted-access Mythos models answer the documented error class.

The Console pages model only credential setup and revocation already required by the customer life. The served API is stateless; Console setup stays local and this pack has no vendor-backed deploy or refresh. The product's core work is its API; this pack adds no chat or generation dashboard. Text token counts and signatures are deterministic local approximations, not a tokenizer or valid model-generated reasoning.