pi-sessions

name	pi-sessions
description	Find, list, and analyse pi agent sessions. Use when the user asks about past pi conversations, session history, costs, what pi did, reviewing a session, or debugging agent behaviour.

Pi Sessions

Pi stores conversation sessions as JSONL files. Each line is a JSON object with a type field.

File Location

~/.pi/agent/sessions/--<path>--/<timestamp>_<uuid>.jsonl

Directory names encode the working directory with / replaced by -.

Example: sessions for /Users/juan.ibiapina/workspace/juanibiapina/gmail-await live in:

~/.pi/agent/sessions/--Users-juan.ibiapina-workspace-juanibiapina-gmail-await--/

Finding Sessions

List all session directories (one per project):

ls ~/.pi/agent/sessions/

List sessions for a project (most recent last):

ls -t ~/.pi/agent/sessions/--<path>--/

Find sessions by date:

find ~/.pi/agent/sessions/ -name "2026-03-18*" -type f

Find sessions containing specific text:

grep -rl "search term" ~/.pi/agent/sessions/--<path>--/

Find the most recent session across all projects:

find ~/.pi/agent/sessions/ -name "*.jsonl" -type f -exec ls -t {} + | head -1

JSONL Record Types

Each line has a type field. Records link via id/parentId forming a tree.

session (header, first line)

{"type":"session","version":3,"id":"uuid","timestamp":"...","cwd":"/path"}

message (conversation turns)

Contains a message object with role:

• user: User input. message.content is string or content blocks.
• assistant: Model response. Has message.content (array of text/thinking/toolCall blocks), message.model, message.provider, message.usage, message.stopReason.
• toolResult: Tool output. Has message.toolCallId, message.toolName, message.content, message.isError.

model_change

{"type":"model_change","provider":"anthropic","modelId":"claude-opus-4-6"}

thinking_level_change

{"type":"thinking_level_change","thinkingLevel":"high"}

compaction

Context was summarized to save tokens.

{"type":"compaction","summary":"...","tokensBefore":50000}

branch_summary

User switched branches in the conversation tree.

custom_message

Extension-injected message (e.g., plan mode).

Useful jq Queries

Session overview (message count, roles, tools used)

cat SESSION.jsonl | jq -r 'select(.type == "message") | .message.role' | sort | uniq -c

List all user messages

cat SESSION.jsonl | jq -r 'select(.type == "message" and .message.role == "user") | .message.content | if type == "string" then . else [.[] | select(.type == "text") | .text] | join("\n") end'

List assistant text responses (no thinking, no tool calls)

cat SESSION.jsonl | jq -r 'select(.type == "message" and .message.role == "assistant") | [.message.content[] | select(.type == "text") | .text] | join("\n") | select(length > 0)'

List all tool calls with names

cat SESSION.jsonl | jq -r 'select(.type == "message" and .message.role == "assistant") | .message.content[] | select(.type == "toolCall") | .name' | sort | uniq -c | sort -rn

List tool calls with arguments

cat SESSION.jsonl | jq 'select(.type == "message" and .message.role == "assistant") | .message.content[] | select(.type == "toolCall") | {name, arguments}'

List tool errors

cat SESSION.jsonl | jq 'select(.type == "message" and .message.role == "toolResult" and .message.isError == true) | {toolName: .message.toolName, content: .message.content}'

Total cost and token usage

cat SESSION.jsonl | jq -s '[.[] | select(.type == "message" and .message.role == "assistant") | .message.usage] | {turns: length, total_input: (map(.input) | add), total_output: (map(.output) | add), total_cache_read: (map(.cacheRead) | add), total_cache_write: (map(.cacheWrite) | add), total_cost: (map(.cost.total) | add)}'

Models used

cat SESSION.jsonl | jq -r 'select(.type == "message" and .message.role == "assistant") | .message.model' | sort | uniq -c

Session duration (first to last timestamp)

cat SESSION.jsonl | jq -r '.timestamp' | sed -n '1p;$p'

Conversation flow (role sequence with timestamps)

cat SESSION.jsonl | jq -r 'select(.type == "message") | "\(.timestamp) \(.message.role)" + (if .message.role == "assistant" then " [\(.message.stopReason)]" elif .message.role == "toolResult" then " [\(.message.toolName)]" else "" end)'

Bash commands executed

cat SESSION.jsonl | jq -r 'select(.type == "message" and .message.role == "assistant") | .message.content[] | select(.type == "toolCall" and .name == "bash") | .arguments.command'

Files read

cat SESSION.jsonl | jq -r 'select(.type == "message" and .message.role == "assistant") | .message.content[] | select(.type == "toolCall" and .name == "read") | .arguments.path'

Files written or edited

cat SESSION.jsonl | jq -r 'select(.type == "message" and .message.role == "assistant") | .message.content[] | select(.type == "toolCall" and (.name == "write" or .name == "edit")) | "\(.name) \(.arguments.path)"'