Use ReadTranscript when agent-transcripts on the box is empty

Grok Bot chat history lives on the server, not in files on the Agent Computer. Empty agent-transcripts/ folders, empty transcript_entries tables, and empty search-index message counts are expected while chats still work in the app. Staff (Colin, Sep 21, 2026) on forum 172494.

Host log lines like server-agent-proxy: server transcript tail dropped … ConnectError are a read-only feed reconnecting after the server closes each stream (~20 minutes by design). Chat delivery does not depend on that feed. Official note that conversations are stored outside the computer: Work with Grok Bot.

Read chat history with ReadTranscript

1. Ask the Bot (or teach the skill / routine) to call ReadTranscript instead of reading local files.
2. With no arguments, ReadTranscript returns that Bot’s own full history — including turns compacted out of the live context.
3. To read another Bot you own, pass agent_id. For another of its conversations use session_id. For a subagent it ran use subagent_id.
4. Results page newest-first. Pass before to walk further back until the routine has what it needs.

Scope

Call	What you get
No args	This Bot’s full history
agent_id	Another Bot on the same account
session_id	A specific conversation of that Bot
subagent_id	A subagent that Bot ran
before	Older page (newest-first paging)

Same-account only: a shared checker can read its owner’s Bots, not a teammate’s.

Pitfalls

• Do not invent or wait for a local agent-transcripts mirror path for forever-box Grok Bot chats — staff say nothing local writes those files for these bots anymore.
• IDE / coding-agent transcript layouts under ~/.cursor/projects/…/agent-transcripts are a different substrate; do not copy those paths into Bot fleet healthchecks.
• The in-app Search palette can still read only a local copy and return nothing for these bots — staff are tracking that separately; use ReadTranscript for healthchecks and routines.