Memory that belongs to your code— not your model
Long-term synthetic memory for developers and teams who don't want to be locked in.
Continuity is a persistent AI memory extension for VS Code, built by Hackerware. It gives AI coding assistants—Claude Code, Cursor, GitHub Copilot, and Gemini CLI—permanent memory across sessions via bounded RAG retrieval stored locally in your project.
Store architectural decisions in your project. Your connected AI tools remember them—across models, IDEs, and teams. Domain profiles adapt vocabulary and templates for software engineering, writing, research, and general knowledge work.
After installing, ask your AI this question:
Your AI should confirm it can access log_decision() and search_decisions(), then summarize the current handoff context when native tools are mounted.
What to Expect Next
- •Handoff context: Connected AI tools can surface repo-local handoff context automatically through Continuity's synthetic memory system; others fall back to repo-local files
- •No manual re-explaining: Once the client is mounted, you can rely on Continuity to surface the relevant project context instead of pasting notes each time
- •Persistent synthetic memory: Decisions you log today remain available to your AI tomorrow, next week, and next month
Benchmarked, Not Promised
Every claim on this page traces to the public whitepaper or measured production data.
vs MemPalace
42 wins, 6 losses, 2 ties across 50 queries • 219× faster wake-up
RAGAS Faithfulness
Faithfulness 0.91 • Answer Relevancy 0.86 • Context Recall 0.86
RGB Robustness
Noise 0.95 • Counter-factual 0.95 • Information Refusal 0.95
vs 3,506ms MemPalace
16ms wake-up • 219× faster than MemPalace (n=50)
Savings at 5K Decisions
~13.8k tokens flat regardless of N • full-context grows linearly to 712k
What is Synthetic Memory?
Context windows are temporary buffers. Synthetic memory is permanent storage for AI.
The Problem: Context Windows Reset
Every new chat starts from zero. You re-explain the same architectural decisions repeatedly.
The Solution: Synthetic Memory
Permanent storage that lives outside the context window—in your project folder.
- •Stored in .continuity/ as plain JSON
- •Works across Claude, Cursor, Copilot, Gemini
- •Local-first, commit to git, share with team
How It Changes Your Workflow
Open any AI chat and it already knows your architecture. No more context-setting.
Why context windows aren't enough
Temporary buffers. You need permanent storage.
Automated Decision Capture
5 detection layers, 19 detection points — capturing decisions automatically.
File System Detection
Monitors 14 architectural file patterns (package.json, tsconfig, Docker, CI/CD) with git-aware diff calculation
Git Hook Integration
Pre-commit hooks detect architectural changes and prompt for decision logging or add to debt tracker. Blocks commit if more than 5 decisions are unlogged.
Memory Middleware
Intercepts AI tool calls to detect 5 decision patterns: research-based, continuity-informed, iterative, config, dependency
Conversation Analysis
Your configured AI provider (Anthropic, OpenAI, Google Gemini, or local Ollama) analyzes conversation logs to find missed decisions — surfaces only results with confidence scoring above 60%. Works offline with template fallback when no provider is set.
Enhanced Prompts
Contextual reminders and accountability metrics shown to AI tools via the synthetic memory system
How Synthetic Memory Works
Log once, remember forever.
Automatic context loading
Every chat starts with your project context already loaded. No copy-paste, no manual prompts.
Works with your AI tools
Claude Code, Cursor, GitHub Copilot, Gemini CLI, and other AI coding assistants. Install once, use everywhere.
Zero-click tool setup
Auto-detects installed AI tools and configures synthetic memory automatically on first activation. No manual config files required.
Visual decision graph (Pro)
See how your architectural choices connect across a 1,500 node capacity knowledge graph. Force-directed, radial, and tree layouts with zoom and filtering.
Dream Engine
4-phase memory consolidation — gathers signals, detects duplicates and contradictions, scores staleness, and consolidates. Like sleep for your project's memory.
Decision lifecycle
Decisions can be active, draft, outdated, deprecated, or superseded. Old decisions are auto-outdated, never deleted — your full history is always preserved.
Local storage only
Everything lives in a .continuity folder in your project. No cloud, no external servers.
Plain JSON format
Your decisions are stored as readable JSON. Commit to git, grep them, edit manually—you own the data.
Domain profiles
Domain profiles adapt Continuity's vocabulary and templates to your field — software engineering, writing, research, or general knowledge work.
Wiki system
Automated wiki health checks detect contradictions, stale claims, and knowledge gaps. Generate summary, comparison, timeline, and deep-dive reports from your decisions.
Anti-sycophancy guardrails
Epistemic Rigor Scoring detects when decisions lack counterarguments. Echo Chamber Detection surfaces groupthink. The only AI memory tool that fights confirmation bias.
CLI feature parity
7 CLI commands — update, archive, lint, report, source, audit, echo-chamber — give full feature access from the terminal without needing a connected AI tool.