Cross-Session Agent Memory

Helps coding agents remember project pitfalls across sessions and retrieve experience on demand

Problem

Developers using Codex and Claude Code find that agents don't remember. Root causes and solutions from a full afternoon of debugging last week are forgotten in a new session, forcing the agent to guess again and even re-propose rejected solutions. Writing notes into AGENTS.md bloats the file, and full injection each session drowns out the relevant bits. Client-native memory is opaque and uncontrollable.

Solution

A lightweight local memory store where the agent automatically drafts candidate lessons when a solution is decided or a root cause is found, and the user confirms what to keep. In new sessions, the agent retrieves relevant lessons on demand based on context instead of loading everything. Entries support tags, search, and export, and integrate with multiple agent clients.

Users

Individual developers and small teams using AI coding agents like Codex and Claude Code.

MVP

Provide a CLI and a simple local database; extract candidate lessons from agent conversations and let users save with one click; expose a query interface for agents to retrieve by keywords or project path; export entries as Markdown or JSON; runs locally with no cloud dependency.

Revenue

Open-source core + paid managed sync (cross-device/team sharing of the memory store); or per-seat team subscription.

Why now

AI coding agents are ubiquitous, but session memory remains a pain point; local vector search and lightweight databases are mature enough for indie developers to build quickly.