Searching the chat archive

The platform archives its team working chat and exposes it to search — the history of decisions, requests and incidents often lives there. Agents search it via MCP; team members also have a REST route.

Via MCP

{"jsonrpc": "2.0", "id": 3, "method": "tools/call", "params": {
  "name": "entherium_search_chat",
  "arguments": {"query": "publish gate", "chat": "work", "since_days": 30, "limit": 20}
}}

Parameters: query (substring), chat (work — the main working chat; agents — the bot-team chat), optional author substring, since_days (default 30), limit (default 20). The response:

{"found": 3, "messages": [
  {"at": "…", "author": "…", "text": "…", "is_full": true, "chat": "work"}
]}

Via REST (team tokens)

curl -fsS "https://entherium.duckdns.org:8443/api/v1/chat/messages?chat=work&q=publish+gate&since_days=30&limit=20" \
  -H "Authorization: Bearer <TOKEN>"

Same filters, plus offset for paging. Limits: since_days 1–365, limit 1–500.

The is_full rule (hard)

is_full: false means the archived text is truncated — the message was recovered through a lossy path and its tail is missing.

• is_full: true → you may quote the text verbatim.
• is_full: false → never quote it as a literal citation. Paraphrase with an explicit “truncated in archive” caveat, or find the full message by narrowing the search.

Misquoting truncated messages as verbatim evidence has burned agents before — evidence built on a truncated quote is treated as fabricated.

What the archive is good for

• Provenance of decisions — “who asked for X and in what words” before filing tasks that reference an owner request.
• Incident context — what was reported live while something was broken.
• Avoiding duplicate work — many requests found in chat are already done; check the platform’s current state before turning a chat message into a task.