Where Kimi Code stores session history

Kimi Code groups sessions by working directory, then gives each session its own folder:

$KIMI_CODE_HOME/sessions/wd_<name>_<hash>/session_<uuid>/

KIMI_CODE_HOME defaults to ~/.kimi-code. There is also a flat index at the root of that directory:

~/.kimi-code/session_index.jsonl

The wd_ group directory

The wd_ prefix stands for working directory. The name that follows is the project folder's own name plus a short hash of its full path:

~/.kimi-code/sessions/wd_blog_dbc648dfdf98/

The hash keeps two projects named blog in different parents from landing in the same bucket, so the directory listing is enough to tell projects apart.

What is in a Kimi Code session directory

state.json                 ← session metadata
agents/main/wire.jsonl     ← the transcript
logs/
media/

wire.jsonl under agents/main/ is the primary transcript. The agents/ level exists because a session can run more than one agent, and main is the one you talked to.

Reading a Kimi Code session from the terminal

List your sessions with their titles, newest first:

jq -r '[.updated_at, .title] | @tsv' ~/.kimi-code/session_index.jsonl | sort -r

Print a transcript:

jq -r 'select(.type) | .text // .content // empty' \
  ~/.kimi-code/sessions/wd_blog_*/session_<uuid>/agents/main/wire.jsonl

Inspect one session's metadata:

jq . ~/.kimi-code/sessions/wd_blog_*/session_<uuid>/state.json

Or open it in an app

Sessions Viewer reads wire.jsonl into the same conversation view as every other agent, keeps the media/ attachments inline, and treats the session directory as the unit for renaming and deleting. It also reads Claude Code, Codex, Grok Build, Pi, Antigravity CLI and opencode.