chipfighter/context-attachment-manager
Turn Codex and Claude Code into your AI assistant director. Pair them with your favorite image-generation tools, keep image-heavy chats lightweight, and find the right image from a rough description—even after repeated context compaction.
What it solves
Long image-based chats with AI coding assistants often slow to a crawl or fail entirely, because every message resends every image in the conversation. This plugin solves that by keeping the images out of the message stream unless the model actually needs them, so requests stay tiny and chats keep going.
How it works
The plugin adds a side panel to Codex and Claude Desktop. Images you generate or share are stored, and the model itself decides which earlier images to pull back into the conversation when it needs them. You can pin important images so they're always sent, or switch to manual selection. It also remembers descriptions of each image (what's in it and what you said) so you can search by memory — type "best face" and it finds that close-up from three turns ago. Everything runs locally; nothing is sent beyond the normal requests.
Who it's for
Developers and creative users who iterate on images inside Codex or Claude Desktop and run into long-chat slowdowns, reconnects, or huge payloads. Also useful for anyone who wants to find past images by remembering something vague about them.
Highlights
- The model manages its own image attachments, pulling back only what it needs.
- Pin images to send every turn, or turn off auto-selection to choose manually.
- Search for any image by what you remember; filter by place or browse by turn.
- Keeps working even after chat compaction.
- In tests, a 42 MB request dropped to 0.12 MB, and a 24.9 MB request went out at 212 KB.
- Available for Codex and Claude Desktop across Windows, macOS, and Linux.
- MIT licensed, with a changelog and contributing guide.
Related
- Project
- Project
- Project
- Project