pond: search across AI agent sessions
pond
Every Claude Code, Codex and other agent session in one search: lossless storage you own, with MCP, CLI, HTTP and SQL access
- 91 ★ on the repository
- Last commit
- Apache-2.0
Extended access
Why this level
- Session history holds code, paths and sometimes keys an agent pasted into a conversation
- An S3 storage bucket must be locked down with your own access rules
Install
- Works in
Manual install
brew install tenequm/tap/pond && pond initmacOS and Linux. Windows has Scoop and prebuilt release archives. Then run `pond sync`.
Access and checks
How we reviewExtended accessThe tool runs code or makes network calls and only reads external data.
- Session history holds code, paths and sometimes keys an agent pasted into a conversation
- An S3 storage bucket must be locked down with your own access rules
This is third-party code. Review the repository files before installing.
What it does
pond collects session history from Claude Code, Codex and other tools, stores it losslessly in a directory or your own S3 bucket, and provides full-text search and SQL queries over it. "How did we fix this last time" becomes an index query with no digging through logs. An MCP server hands the data to agents: the tools `pond_search`, `pond_get_session`, `pond_get_message`, `pond_ingest` and the read-only `pond_sql`. The author reports a measurement: a ranked answer takes 1.5 to 3 thousand tokens, while grep plus reading a matching transcript costs 90 to 255 thousand. A session can be restored into any supported client and continued there. The `pond init` command sets up storage, adapters and a schedule, registers the MCP for Claude Code and installs a bundled skill. `pond sync` ingests and indexes new sessions.
Who it is for. Developers who work with several agents on different machines and want to search their own history for past decisions and fixes.
Good fit when
- You remember fixing a similar bug but not which agent or when
- You want to continue a session started in Claude Code inside Codex
- Session history sits on several machines and must be gathered in one S3 bucket
- An agent needs read-only SQL access to its own past conversations
Not a fit when
- Search inside a single agent is enough: Claude Code's built-in history covers it
- You want distilled facts instead of full records: a memory layer fits better
- Disk space is tight: the prebuilt Windows binary takes about 223 MB unpacked
Example request
Check in pond how we fixed the stuck queue three months ago and apply the same fix here.MCP
- Transport
- stdio
- Authentication
- not required
Limitations
Pre-v1 status: schemas, wire shapes and config keys may change incompatibly until the first release. Full documentation is at pond.locker and the contract lives in `docs/spec.md`.
For Codex, `pond init` prints the MCP registration command and you run it by hand. The client must be restarted after registration or the tools will not load. The one-line skill install targets a POSIX shell, and PowerShell 5.1 has a separate block in the docs.
How to disable. Unregister with `claude mcp remove pond` and `codex mcp remove pond`, delete the `~/.claude/skills/pond` directory and remove the package with `brew uninstall pond`. The sessions stay in the chosen storage, which you clear separately.
FAQ
How do I install pond in Claude Code?
Install the package with `brew install tenequm/tap/pond`, then run `pond init` and `pond sync`. Initialization registers the MCP and installs the skill. By hand the same is `claude mcp add -s user pond -- pond mcp`.
Where are the sessions stored?
In a directory you choose or in your own S3 bucket. There is no author-run service between you and the data.
How is pond different from memory tools?
It stores whole sessions and searches them, while memory layers keep a digest of facts. The author maintains a per-tool comparison at pond.locker/compare.
Can a session move between agents?
Yes. Any session can be restored into any supported client and continued there.
Does search work in Russian?
Yes. Full-text parsing is word-level and language-neutral, and tokens without a known stem stay exact matches.
Discussion
No questions or reviews yet.
Related
MCP serversA SQLite archive of Claude Code sessions with an MCP server, a terminal UI and a web UI for finding past conversations.
compound-memory turns your AI chats into memory about you each night and injects it into new Claude Code and Codex sessions
An MCP server for searching your Claude Code conversation history: past solutions, errors, plans and file change history.
The aise utility searches local session history of Claude Code, Codex, Cursor and six more agents and serves it over MCP.
Local browsing, search and cost tracking for Claude Code, Codex and other agent sessions in one web UI
A skill for searching past sessions: a full-text index of Claude Code, Codex, pi and Grok chats in one SQLite database
Comparison with similar tools
Cards that share the main search query, and how they differ.
| Tool | Official | Authentication | Repository stars | Main difference |
|---|---|---|---|---|
| pond: search across AI agent sessions (this card) | no | not required | 91 | |
| Claude-Mem | no | 116,457 | compressed memory as facts, no full sessions | |
| Recall | no | 768 | local search over one agent's transcripts |
README in short
The README states the problem, gives a quick start and compares pond with local session indexes and memory layers. It goes on to installation for several platforms, client adapters and the transports: HTTP with JSON and MCP share the same handlers. A separate section holds token cost measurements on five questions about past sessions.