Anvia

Messages ​

Messages are Anvia's provider-neutral transcript format. Direct completions, agents, memory stores, tools, guardrails, observers, and UI adapters use the same roles and typed content blocks.

Provider adapters translate these normalized messages into their native request formats and translate provider responses back into normalized assistant content.

1. Create a transcript ​

Messages are strict JSON-safe objects. Use the exported types for authoring and parseMessages() at untrusted boundaries:

ts
import { parseMessages, type Message } from '@anvia/core'

const transcript = [
  { role: 'system', content: 'Answer clearly and state uncertainty.' },
  { role: 'user', content: 'Summarize this incident for the support team.' },
] satisfies Message[]

const validated = parseMessages(JSON.parse(requestBody))

Each role has a fixed content shape. Optional metadata must also be strict JSON.

2. Send the transcript ​

The same Message[] can be passed to a direct completion or an agent:

ts
const completion = await generateCompletion({
    messages: transcript,
    model
})

const agentResult = await incidentAgent.generate({
    messages: transcript
})

A direct completion or stateless agent sends the entire array as chat history. A memory-backed run instead uses { prompt, session }; persisted sessions and caller-owned transcripts are mutually exclusive.

3. Use discriminated content parts ​

User content arrays contain text, image, and file parts. Their discriminants keep provider-neutral media explicit:

ts
import type { UserMessage } from '@anvia/core'

const prompt: UserMessage = {
  role: 'user',
  content: [
    { type: 'text', text: 'Describe the attached diagram.' },
    {
      type: 'image',
      image: { type: 'url', url: 'https://files.example.com/architecture.png' },
    },
  ],
}

Anvia validates optional capabilities such as image input and file-document input before calling the provider. The adapter capability is still only the first check: the exact provider model, account, file type, and size limit must also support the request.

Explore messages ​

  • Message roles explains system, user, assistant, and tool messages.
  • Content types builds text, image, document, reasoning, and tool blocks.
  • Documents separates tagged text context from file attachments.
  • Tool call preserves a complete tool exchange.
  • Reasoning handles display text, opaque provider content, and generation metadata.
  • Transcripts continues and persists ordered conversations.

Continue with Message roles.

Built for Anvia.