Local work journal
Records supported tool inputs and results, including useful failures such as timeouts or empty results, while a turn is in progress.
Eggshell
Eggshell is a local memory tool for Codex CLI that records coding-agent work, organizes it on the developer’s machine, and hands relevant findings to later chats. It helps developers continue repository investigations and follow-up tasks without rebuilding context from scratch.
Local memory for AI coding agents
Eggshell is a local memory tool for AI coding agents, centered on Codex CLI. It records work from one chat—such as repository searches, commands, documentation findings, tool results, and conclusions—and makes relevant history available to a separate chat in the same project. Its purpose is to reduce the need to repeat investigation when an agent conversation ends or a follow-up starts.
The Codex integration stores project memory in local Eggshell files and uses local search and graph processing to organize it. A later agent receives a selected handoff and is instructed to reuse supported findings, check facts that may have changed, and identify what it reused or left unverified. The project also includes separate experimental adapters for Claude Code, Gemini CLI, Cursor, and OpenCode, but those integrations are built and installed independently from the published Codex plugin.
Core features
Relevant-history retrieval
Uses local text matching and MiniLM embeddings to find related work in files that the project configuration allows Eggshell to read.
Cross-chat handoffs
Connects requests, operations, and outcomes in a graph, then sends selected prior work as context to a new chat.
Incremental persistence
Saves each observation independently and retries interrupted writes; unfinished work remains open while completed turns add a parent result.
Local organization
Runs search and graph processing locally without using an LLM to summarize memories, classify them, or maintain the graph.
Explicit workflow controls
Provides Codex controls such as `!egg doctor`, `!egg graph`, `!egg keep`, and `!egg drop` for checking setup, inspecting delivery, saving a turn, or clearing the active turn.
Practical use cases
Continuing repository investigations
Record commands, searches, and source findings during a codebase investigation, then ask a new chat to continue from the documented evidence rather than repeat the initial exploration.
Follow-up implementation work
Save the outcome of a completed implementation or debugging turn and pass the relevant decisions and supporting operations to a later task in the same project.
Checking handoff provenance
Use `!egg graph` to inspect which prior work was actually delivered to the agent, helping verify that a handoff contains the expected context before proceeding.
Recovering interrupted sessions
Keep incomplete work available after an interrupted session; queued writes retry automatically and the next task can distinguish historical results from facts that need rechecking.
Testing alternate coding agents
Install one of the separate adapters when working in Claude Code, Gemini CLI, Cursor, or OpenCode, while treating those integrations as experimental rather than equivalent to the documented Codex workflow.
Pros and Cons
Pros
- Keeps memory organization and retrieval local, without LLM calls for summarization or graph maintenance.
- Preserves intermediate tool evidence and unfinished work instead of saving only a final summary.
- Supports project-level cross-chat handoffs so a follow-up can build on earlier investigation.
- Provides explicit controls for inspecting delivery and managing the active turn.
Cons
- The primary documented integration requires Codex CLI plugins and command hooks; ordinary ChatGPT Chat does not provide automatic memory.
- The Claude Code, Gemini CLI, Cursor, and OpenCode adapters are experimental, source-built integrations, and their live-session token savings have not been evaluated.
- Selected memory and subsequent agent work still consume the model's normal input and output tokens.
FAQ
What platforms and prerequisites does Eggshell support?
Eggshell is designed for Codex CLI projects on macOS or Linux, with Apple Silicon/ARM64 or x86-64 support. The Codex client must support plugins and command hooks.
How do I install and verify Eggshell?
Install the Codex plugin with the documented installer or from a source checkout, then run `egg init` in the project. In Codex, review and enable Eggshell's hooks through `/hooks`, start a new chat, and use `!egg doctor` to check the setup.
How does memory move between chats?
After initialization, Eggshell records supported tool results and final answers locally, searches relevant prior work, and sends selected history as a handoff to a later chat. Ordinary tasks do not require special prompts, although `!egg keep` can explicitly save a finished turn.
Can Eggshell be used with other AI coding clients?
The Codex integration uses command hooks, so ordinary ChatGPT Chat does not provide automatic Eggshell memory. The adapters for Claude Code, Gemini CLI, Cursor, and OpenCode are separate experimental, source-built integrations.
Does Eggshell eliminate model-token usage?
Eggshell's search and graph processing run locally without LLM calls for summarization, classification, or graph maintenance. Retrieved memory and the agent's later work still use the model's normal input and output tokens.
Quick Facts
- Category
- Developer Tool
- Primary client
- Codex CLI
- Deployment model
- Local project memory and processing
- Supported primary platforms
- macOS and Linux; Apple Silicon/ARM64 or x86-64
- Project source
- github.com/momonpya/eggshell
- Other integrations
- Experimental adapters for Claude Code, Gemini CLI, Cursor, and OpenCode
Eggshell Alternativen
ByteAsk
ByteAsk is a terminal-first AI coding agent for C and C++ that edits repositories and verifies changes with the real compiler, debugger, sanitizers, and tests before showing a diff. It offers a free tier plus paid plans, with editor connectors and zero-retention handling described in the source.
Goldfish
Goldfish is an AI memory app for macOS that helps you reply, write, summarize, and continue using the context already on your Mac. It stores memory locally and writes in your tone.
Lasso
Lasso is an ecommerce product data platform for enriching catalog records, processing supplier files, generating product content, and monitoring competitors. It combines a web app with a REST API, SDK, and MCP server for teams and developers.
CreateOS Sandbox
CreateOS Sandbox is an isolated compute environment for running code and agent workloads inside Firecracker micro-VMs. It is designed for workflows that need machine-level isolation, private networking between sandboxes, and programmatic control through SDK, CLI, or MCP.
ArtDeck
ArtDeck is a visual reference board app for iPhone, iPad, and Mac that lets artists collect mixed reference material on one canvas and study it with ToolBox lenses. It works offline, supports optional iCloud sync, and is sold as a one-time purchase with no account required.
GitBot
GitBot is a local browser-based workspace for creating reusable coding bots powered by Claude Code, Codex, or OpenCode. It helps developers run those agents across repository folders with persistent threads and configurable permissions.