Self-evolving Context Database for AI Agents. Unify Agent Memory, Knowledge RAG and Skills.
Python agent-contextagent-memoryagent-pluginsagentic-ragcontext-databasedsh-pluginknowledge-basememorymemory-managementself-evolvingvector-database
import type { Message, MessagePart } from './types/message'
/** Presentation only: retain server records and associate tool transports within a turn. */
export function groupMessages(messages: Message[]): Message[] {
const groups: Message[] = []
for (const message of messages) {
const parts = (Array.isArray(message.parts) ? message.parts : []).filter(
(part) => part.type !== 'text' || Boolean(part.text.trim()),
)
if (!parts.length) continue
const transport =
message.message_kind === 'tool_transport' ||
(!message.message_kind &&
message.role === 'user' &&
parts.every(
(part) => part.type === 'tool' || part.type === 'tool_result',
))
const role = transport ? 'assistant' : message.role
const previous = groups.at(-1)
const sameTurn =
!previous?.turn_id ||
!message.turn_id ||
previous.turn_id === message.turn_id
if (
role === 'assistant' &&
previous?.role === 'assistant' &&
sameTurn &&
message.message_kind !== 'checkpoint' &&
previous.message_kind !== 'checkpoint'
) {
previous.parts = mergeParts(previous.parts, parts)
previous.turn_id ??= message.turn_id
} else {
groups.push({ ...message, role, parts: mergeParts([], parts) })
}
}
return groups
}
function mergeParts(
existing: MessagePart[],
incoming: MessagePart[],
): MessagePart[] {
const parts = [...existing]
for (const part of incoming) {
if (part.type === 'tool' && part.tool_id) {
const index = parts.findIndex(
(candidate) =>
candidate.type === 'tool' && candidate.tool_id === part.tool_id,
)
if (index >= 0) {
const previous = parts[index]
if (previous.type === 'tool')
parts[index] = {
...previous,
...part,
tool_input: part.tool_input ?? previous.tool_input,
tool_output: part.tool_output || previous.tool_output,
}
continue
}
}
parts.push(part)
}
return parts
}