Evenfire is an intelligence layer platform for running LLM agents that take real actions, on your infrastructure
TypeScript
import type {
ChatMessage,
ChatMessageAttachment,
MessageToolStep,
SessionTokensLite,
} from './types'
export interface ServerTurn {
number: number
user_input: string
response?: string
started_at: string
completed_at?: string
tokens?: SessionTokensLite
tool_steps?: MessageToolStep[]
}
type UserMessageDisplay = {
content: string
attachments: ChatMessageAttachment[]
}
/**
* Convert runtime turns to deterministic local messages.
*
* The optional display parser lets the renderer recover structured attachment
* chips while keeping server-turn identity shared with main-process tests.
*/
export function turnsToChatMessages(
turns: ServerTurn[],
parseUserDisplay: (content: string) => UserMessageDisplay = content => ({
content,
attachments: [],
})
): ChatMessage[] {
const out: ChatMessage[] = []
for (const turn of turns) {
const userDisplay = parseUserDisplay(turn.user_input)
out.push({
id: `turn-${turn.number}-user`,
role: 'user',
content: userDisplay.content,
timestamp: Date.parse(turn.started_at),
serverTurnNumber: turn.number,
...(userDisplay.attachments.length ? { attachments: userDisplay.attachments } : {}),
})
if (turn.response !== undefined) {
out.push({
id: `turn-${turn.number}-assistant`,
role: 'assistant',
content: turn.response,
timestamp: Date.parse(turn.completed_at ?? turn.started_at),
serverTurnNumber: turn.number,
...(turn.tokens ? { tokens: turn.tokens } : {}),
...(turn.tool_steps?.length ? { toolSteps: turn.tool_steps } : {}),
})
}
}
return out
}