Athena-Public: Keep AI Coding Agent Context Across Sessions

Overview

Project Athena is a Python tool for maintaining project and session context for AI coding agents. It stores memory as Markdown files on disk, then uses /start and /end workflows to load prior context and record what changed, addressing the repeated setup that happens when an agent starts a new session without memory.

It is most relevant to developers who work across coding agents or IDEs and want their context to remain under their control. Search can combine keyword and optional vector retrieval, while portability and enforcement vary by IDE: Claude Code has hooks, whereas several other integrations rely on instructions to the agent.

Key Features

• Creates a workspace of Markdown memory files for user preferences, product context, active work, patterns, and session logs.
• Provides /start and /end routines to load a checkpoint and save a session summary.
• Generates IDE-specific instruction files through athena init --ide for supported targets.
• Offers hybrid keyword and optional vector search for finding relevant memories.
• Includes Claude Code hooks for checks such as secret scanning and destructive-command safeguards.
• Supports optional Supabase-based cloud sync and reranking with the full install.

Use Cases

• A developer moving between Claude Code, Cursor, or another supported IDE can keep project context in files rather than relying solely on one tool's chat history.
• A solo developer returning to a project after a break can use session checkpoints to restore what was in progress.
• A developer who prefers local, versionable context can keep memory as Markdown and choose whether to add cloud-backed search.
• A team or individual evaluating agent memory can inspect and edit the stored context directly instead of treating it as hidden platform state.

Getting Started

git clone https://github.com/winstonkoh87/Athena-Public.git && cd Athena-Public
python3 -m venv .venv && source .venv/bin/activate
pip install -e ".[local]"
athena init --ide claude

Then use /start and /end in the IDE's AI chat panel. See the README and Getting Started guide for configuration and other IDE targets.

Alternatives

• ReMe: ReMe makes agent conversations and resources searchable, editable Markdown memory, while Athena centers on carrying coding context between sessions.
• cabinet: Cabinet pairs a local Markdown knowledge base with agents, scheduled jobs, and team workflows; Athena focuses on a portable coding-session routine.
• engram: Engram stores agent memories in local SQLite and exposes them through MCP, a CLI, and APIs rather than relying on Markdown files.

Considerations

• IDE support is not uniform. The README marks some integration load paths as in progress or untested, and the listed guardrail hooks are implemented for Claude Code only.
• Outside Claude Code, guardrails are instructions for the model rather than code-enforced checks.
• The README describes compounding checkpoints and canonical fact updates as prompt-level behavior, with a code-enforced writer still in development.
• The local install is the lighter option. Cloud sync and reranking require the full install, and the README notes that cloud SDKs are currently included in base dependencies.