Continuity - Synthetic Memory for AI Coding Assistants Logo
Continuityby Hackerware
Start
Tutorial 1 of 714% complete
All tutorials
5 minBeginner

Quick Start Guide

Set up AI coding assistants and fallback paths

Quick Start Guide

Get Continuity's RAG-based synthetic memory system running in 5 minutes.

Setup

bash
# 1. Install from VS Code Marketplace
#    Search "Continuity" → Install

# 2. Open any project folder in VS Code
#    Continuity automatically configures communication with your AI assistants.

# 3. Restart your AI tool if prompted
#    (Claude Desktop, Claude Code, Cursor, Cline, Roo Code, GitHub Copilot, or Gemini CLI)

Continuity provides a unified RAG-based memory system that works across multiple AI coding tools. It uses standard protocols like MCP when the host/runtime supports native tool mounting, ensuring seamless interaction with your decisions and context. If native tool mounting is unavailable, Continuity falls back to repo-local files for context preservation.

Tip: If your AI assistant doesn't automatically detect the memory system, run Cmd+Shift+P → "Continuity: Configure Memory" to trigger it manually.

What Happens Automatically

On first launch, Continuity can run the Initial Project Scanner — it scans your codebase and creates draft decisions for your tech stack, dependencies, database, testing setup, deployment config, and project structure. Combined with git seeding, which pre-populates decisions from your recent git history, your AI begins with a robust synthetic memory instead of an empty slate.

Continuity also creates instruction files (CLAUDE.md, .cursorrules, .github/copilot-instructions.md, and others depending on your AI tools) that your AI can read on startup. These files act as pointers to the synthetic memory system, telling your AI:

  • •When to search for existing decisions
  • •When to log new decisions immediately
  • •How to check for conflicts with past choices

When you ask an architectural question, your AI will search its synthetic memory to find the most relevant prior decisions.

When you explain your reasoning, your AI will immediately record it as part of the project's permanent synthetic memory.

See It In Action

Ask an architectural question

You: "Should we use Redis or Memcached for caching?"

AI: Searches for related decisions

"Let me check what we've decided about caching... I found decision-23 where you chose PostgreSQL and prioritized reliability. Based on that pattern, Redis might be a better fit since it offers persistence. What do you think?"

Explain a choice

You: "Let's use TypeScript because the type safety catches bugs early and the IDE support is better."

AI: "Good reasoning. I'll log this decision right away so we remember why TypeScript when it comes up later."

Check for conflicts

You: "I'm thinking of switching to MongoDB."

AI: Searches for database decisions

"Before we decide, I found decision-12 where you chose PostgreSQL specifically for ACID transactions. Switching to MongoDB would change that approach. Do you want to proceed, or should we look at what drove that original decision?"

If Your AI Doesn't Behave This Way

Some AI tools don't reliably read instruction files. If your AI isn't being proactive:

  1. •Ask explicitly: "Search for decisions about caching before we decide"
  2. •Remind it: "Check the CLAUDE.md file for how to use Continuity"
  3. •Direct request: "Log this decision: We chose X because Y"

The tools work when asked directly on clients that expose them. If they are not mounted, use the repo-local fallback files.

Key Commands

You can always ask your AI to:

What you wantWhat to say
Search decisions"What have we decided about X?"
Log a decision"Log this decision: [your reasoning]"
Load context"What's the project context?"
Track a blocker"I'm stuck on X, add it to blockers"

Troubleshooting

AI not using Continuity memory?

  • •Continuity auto-detects your AI assistants on startup. If it missed one, run Cmd+Shift+P → "Continuity: Configure Memory" to re-run detection
  • •Restart your AI tool completely after setup
  • •Check that the instruction file for your AI exists (CLAUDE.md for Claude Code, .cursorrules for Cursor, etc.)
  • •If the client is in degraded mode, inspect .continuity/mcp-health.json and .continuity/unfinished-task.json

AI being proactive but missing context?

  • •Decisions are stored in the synthetic memory system (.continuity/decisions.json)
  • •Run "Continuity: Regenerate Instruction Files" to update the AI's pointers to memory

Next Steps

Fluent AI Memory →