logsesh

Local-first CLI for AI coding-agent session logs.

Search past work, inspect usage, estimate cost, and export sessions from Claude Code, Codex, and Antigravity CLI logs. No telemetry. No runtime network calls.

npm install -g logsesh
logsesh doctor

Requires Node.js >=22.

Screenshots use mock data.

Use

logsesh list
logsesh search 'project:myapp auth middleware'
logsesh stats --since 7d --estimate-cost
logsesh export --format markdown --since 30d --out sessions.md

doctor is the best first command on a new machine. It checks log discovery, adapter health, export defaults, and bundled pricing data.

Commands

command	purpose
doctor	check log access, adapters, format health, and pricing data
list	show matching sessions
search	find transcript text with snippets
stats	summarize activity, tokens, and cost
export	write JSON, JSONL, Markdown, or CSV
index build	build the optional local SQLite search index
index status	show index health
index clear	delete the local index
index path	print the index database path

Run logsesh <command> --help for all options.

Output

Human output is rich by default when stdout is a TTY. Use --json for scripts and --plain for stable text.

logsesh stats --since 7d         # rich dashboard
logsesh stats --since 7d --json  # machine-readable envelope
logsesh stats --since 7d --plain # stable text

Query

The same query syntax works across search, list, stats, and export.

query	meaning
auth middleware	match transcript text
auth AND middleware	require both terms
"rate limit"	exact phrase
project:myapp	match project directory name or path segment
tool:codex	match a tool
model:gpt-5.6	match a model id
agent:Explore	match agent/subagent id or type
parent:session-id	match parent session id
branch:feat/x	match git branch
toolcall:Bash	match a tool-call name
project:myapp auth	project filter plus text search

--project myapp is shorthand for project:myapp.

list, search, and stats use a local SQLite index when one exists. --no-index and --roots always scan source files. --large-files auto|degraded|full controls files over 512 MiB; pass --max-file-bytes to restore the old skip-the-file cap.

Safety

Transcript exports redact known sensitive patterns by default. Paths are anonymized unless you opt out.

flag	effect
--allow-sensitive / --no-redact	export full transcript text
--redact-pattern <regex>	add a custom redaction pattern
--include-reasoning	include reasoning blocks
--no-anonymize-paths	keep raw filesystem paths
--unsafe-raw	disable Markdown injection neutralization

Summary-only CSV keeps path anonymization on by default. Pass --redact to apply pattern redaction to summary rows too.

Supported Logs

tool	default location
Claude Code	~/.claude/projects/*/*.jsonl and <sessionId>/subagents/agent-*.jsonl
Codex	~/.codex/sessions/**/rollout-*.jsonl and ~/.codex/archived_sessions/**
Antigravity CLI	~/.gemini/antigravity-cli/brain/<id>/.system_generated/logs/transcript_full.jsonl, conversations/<id>.db, plus ~/.gemini/antigravity and ~/.gemini/antigravity-ide. Other ACP stores: --roots antigravity:<path>
Gemini CLI (legacy)	~/.gemini/tmp/*/chats/session-*.jsonl

Override discovery with --roots tool:path and repeat it for multiple roots.

Library

npm install @logsesh/core

import { runPipeline, sanitizeForExport } from "@logsesh/core";

for await (const { session, warnings } of runPipeline({ toolFilter: ["codex"] })) {
  for (const warning of warnings) console.warn(warning.message);
  if (!session) continue;

  console.log(sanitizeForExport(session));
}

Schemas are published under @logsesh/core/schemas/*. Pricing data is published under @logsesh/core/pricing/*.

License

MIT