Skip to content

CLI · AI session export

Export your AI coding sessions.

Export local conversations from Codex, Claude Code, and other AI coding tools to Markdown or JSON. Search your session history from one CLI.

Export a Codex session to Markdown

Seven tools, one URI grammar

Every session is addressable by its provider scheme. Point agent-dump at a URI and read it anywhere.

  • Codexcodex://threads/a1b2c3
  • Claude Codeclaude://a1b2c3d4
  • ZCodezcode://sess-9f8e7d
  • Kimikimi://k-7a3f21
  • OpenCodeopencode://5c4b3a
  • Cursorcursor://req-8821
  • Pipi://019e7978-b2ec
  • More toolsPRs welcome

What it does

Agent Dump gives developers one command-line interface for local AI coding session history across seven tools. It turns provider-specific session stores into readable exports, direct URI views, search results, stats, and collection reports.

Reads every local session

Codex, Claude Code, ZCode, Kimi, OpenCode, Cursor, and Pi, from their native stores.

agent-dump --interactive

Searches and filters

Full-text search across titles, messages, and reasoning; filter by provider, role, or path.

agent-dump --search "auth timeout"

Collects summaries

High-signal session digests for project management and insight reports.

agent-dump --collect

Install

Works with uv, npm, pnpm, and bun. JavaScript package wrappers require Node.js 22+.

Install globally

uv tool install agent-dump

Run without installing

uvx agent-dump --help

Or add it as an agent skill

npx skills add xingkaixin/agent-dump

What's New

Continuous evolution towards seamless multi-agent coding workflows, deep search, and high-signal session memory.

Export and Search DeepChat and Cherry Studio Sessions

Bring saved DeepChat conversations and Cherry Studio 2.x chats and agent sessions into your AI session export, full-text search, and AI collect workflows. Export Markdown or JSON while keeping source data unchanged. New English and Chinese guides walk you through exporting Codex sessions.

agent-dump --list -days 30 -query "provider:deepchat,cherry"

More Reliable Collect and Up-to-Date Session Reads

Keep more visible dialogue in AI collect summaries, apply query filters consistently, and see when unreadable sessions leave gaps. OpenCode and ZCode session reads and full-text search now refresh after database changes. The landing page also loads faster.

agent-dump --collect -days 7 -query "provider:codex path:. limit:20" --dry-run

Resilient Collect, Provider Pre-Scoping & Visual Redesign

Collect workflows now gracefully preserve partial session successes during batch failures and isolate unknown projects cleanly. Scope providers upfront before discovery to accelerate queries, paired with an all-new WebGL landing visual experience.

agent-dump --collect --agent codex --days 7

External Agent Handoff & Visible Dialogue Summaries

Generate self-contained prompts with safe read-only URI commands using `--emit-prompt`, allowing external AI agents to autonomously collect sessions and author reports without local API configuration. Session summarization now strictly filters for visible human/assistant dialogue.

agent-dump --collect --emit-prompt --save ./reports/weekly.md

Full-Text Search & Unified Relevance Ranking

Fast SQLite FTS5 full-text indexing with intelligent CJK boundary segmentation. Search titles, transcripts, and reasoning across all seven AI coding tools with unified corpus scoring and literal phrase matching.

agent-dump --search "auth timeout" --days 30

FAQ