kapsel（kapsel 开发）

Kapsel — Project Memory Capsules

Kapseln (capsules) let an AI agent archive everything it knows about a completed project into a structured folder on cloud storage. When the project is needed again, the agent loads the capsule and has full context — without carrying dead project knowledge in its permanent memory.

Think of it like moving a finished project from your desk into a labeled filing cabinet. Your desk stays clean, but you can pull the folder out anytime.

How it works

Each capsule is a folder on a cloud remote (via rclone) with this structure:

<remote>:<base-path>/Kapseln/<project-name>/
├── summary.md    — Short overview (always readable, max 500 words)
├── details.md    — Decisions, timeline, links, background
├── context.md    — Technical details, configs, code snippets
└── files/        — Any associated files (optional)

The summary.md is deliberately kept short so the agent can scan all capsules quickly and decide which one to load in full.

Setup (one-time)

The script needs rclone configured with at least one cloud remote. If the user hasn't set up rclone yet, guide them through it:

rclone config    # Interactive setup wizard

After rclone is configured, the user needs to set two things in the script or via environment variables:

Set them as environment variables or edit the top of kapsel.py.

export KAPSEL_REMOTE="gdrive:MyAgent/Kapseln"
export KAPSEL_TMP="/tmp/kapseln"

Commands

Run the script from the workspace scripts directory:

python3 scripts/kapsel.py list                    # Show all capsules with summaries
python3 scripts/kapsel.py create <name>           # Create new capsule (empty template)
python3 scripts/kapsel.py load <name>             # Load full capsule (all docs)
python3 scripts/kapsel.py summary <name>          # Show only the short summary
python3 scripts/kapsel.py archive <name>          # Mark as completed
python3 scripts/kapsel.py save <name> <file>      # Add a file to the capsule

When to use each command

Starting a new project — create makes an empty capsule with template files. Fill in the summary, details, and context as the project progresses.

Project is done — archive marks the capsule as completed. After archiving, you can safely forget the project details from your active memory. The capsule preserves everything.

Need old project knowledge — summary gives a quick refresher. If you need the full picture, use load to get all details and technical context.

Want to store a file — save copies any file into the capsule's files/ folder. Use this for configs, exports, screenshots, or any artifact worth keeping.

Workflow for the agent

The recommended pattern for an AI agent using capsules:

1. When a new project begins: kapsel.py create my-project
2. As work progresses: update the capsule files with learnings, decisions, configs
3. Project completed: kapsel.py archive my-project
4. Remove project details from active memory (e.g. MEMORY.md) — the capsule is the archive
5. Later, if the project comes up again: kapsel.py load my-project

The key insight is that capsules free up the agent's working memory. Instead of accumulating ever-growing context about every project, the agent keeps only active projects in memory and offloads completed ones to capsules.

Installation

1. Copy scripts/kapsel.py into your agent's workspace scripts directory
2. Make sure rclone is installed and configured with a cloud remote
3. Set KAPSEL_REMOTE to your preferred storage path
4. Add the commands to your agent's memory/instructions so it knows they exist