Build your own local Continuity Layer
thecontinuitylayer.com/build-your-continuity-layerA simple local setup for carrying project state, decisions, handoffs, and progress across AI sessions. The runtime is volatile. Continuity must be durable.
How to use this page with an AI assistant
This guide is meant to work whether your assistant can browse the web or only sees what you paste into the chat.
- 01If your AI can browse, paste this page link and ask it to guide you through the setup one step at a time. Do not assume every AI can browse links or write local files; ask it what it can safely do first.
- 02If your AI cannot browse, copy the starter prompt below and paste the folder structure and packet template into the same chat.
- 03Ask the AI to proceed one step at a time. A continuity layer improves through review, not by stacking a dozen unreviewed prompts in a row.
Copyable starter prompt
Use this when your assistant cannot browse the page, or when you want a clean setup prompt in a fresh session.
Help me set up a local Continuity Layer for my project. Use this guide as the source of truth. Start simple: - local files first - backup-first - manual imports only - no cloud sync - no secrets - no auto-import - one step at a time Help me create the folder structure, then help me make my first continuity packet. Ask me only one or two questions at a time.
Minimal local folder structure
The folder is the starting point. Keep it local, readable, and boring enough that you can audit it later.
continuity-layer/ memory/ imports/ exports/ decisions/ handoffs/ backups/ ACTIVITY_LOG.md CURRENT_STATE.md SETUP_JOURNEY.md
- memory/
- Durable notes about the project: current facts, standing constraints, and reviewed context.
- imports/
- Human-approved material waiting to be folded into the continuity layer.
- exports/
- Packets you hand to an AI assistant at the start of a new session.
- decisions/
- Dated records of choices that should carry forward until superseded.
- handoffs/
- Planner-to-builder or session-to-session prompts, outputs, and review notes.
- backups/
- Copies made before deleting, reorganizing, importing, or rewriting durable files.
- ACTIVITY_LOG.md
- A chronological record of what happened, what changed, and what was checked.
- CURRENT_STATE.md
- The maintained answer to where the project stands now.
- SETUP_JOURNEY.md
- Notes on how the local continuity layer itself was created and improved.
First continuity packet
A packet is the compact handoff you give the next session so it can start from reviewed state instead of raw transcript.
- Project: Fill this in after review.
- Current goal: Fill this in after review.
- Current state: Fill this in after review.
- Decisions made: Fill this in after review.
- Active constraints: Fill this in after review.
- Source material: Fill this in after review.
- Next action: Fill this in after review.
- Parking lot: Fill this in after review.
- Open questions: Fill this in after review.
- Generated at: Fill this in after review.
Planner / Builder handoff loop
Keep the loop small enough that the human can review it.
- 01Planner gives one prompt.
- 02Builder works locally.
- 03Builder reports output.
- 04Human reviews.
- 05Planner decides the next prompt.
- 06Do not stack prompts.
What makes this a continuity layer
Notes store information. Continuity governs what carries forward. A local Continuity Layer keeps the current state, decisions, provenance, supersession, handoffs, blocked actions, and next action visible enough that a new session can continue the work without pretending the whole transcript is equally current.
Layer, engine, runtime
- The Layer
- The responsibility: carrying project state, memory, provenance, decisions, identity, and progress forward.
- The Engine
- The implementation: files, rules, scripts, or databases that apply continuity rules and maintain the current account.
- The Runtime
- Where live interaction happens: the chat, editor, agent, or tool session using the continuity layer.
Later: turning it into an engine
Once the manual workflow is reliable, you can add machinery. None of this is required for day one.
- SQLite or another local database
- import/export scripts
- validation checks
- provenance fields
- supersession chains
- packet generation
Safety and boundaries
Make the local workflow trustworthy before you make it powerful.
- Do not put secrets, API keys, passwords, or private credentials into memory files.
- Keep it local first.
- Back up before deleting, reorganizing, importing, or rewriting durable files.
- Avoid auto-import until the manual workflow is reliable.
- Treat AI output as draft state until reviewed.