Continuity — Repo-Local Context for AI Coding

Every AI session starts from zero. Your assistant forgets the database you chose last week, the pattern you rejected yesterday, and the constraint your team agreed on this morning.

Continuity fixes this. It helps MCP-capable AI tools recover structured, repo-local context instead of starting cold every time.

---

How It Works

1. Log decisions as you work — manually, via MCP, or auto-detected
2. AI recalls them across sessions via MCP integration
3. Decisions live in your repo — .continuity/decisions.json, committed alongside your code

No cloud. No vendor memory. You own the state.

---

Core Features

• Repo-Local Decision History — Log architectural decisions once; connected AI tools can recover the why across sessions
• MCP Integration — Claude Code, Cursor, Cline, Copilot, Gemini, and any MCP-compatible tool
• Auto-Detection — Watches for significant code changes and prompts you to log them
• Semantic Search — Find decisions by meaning, not just keywords
• Code Intelligence — Link decisions to specific files and symbols
• Session Handoff — Auto-generate context documents with delta tracking
• Knowledge Graph — Visualize decision relationships in an interactive graph
• MCP Setup Wizard — One-click configuration for Claude Code, Cursor, Cline, and more

---

Getting Started

1. Install from the VS Code Marketplace
2. Open a project folder in VS Code
3. Configure MCP — Command Palette → Continuity: Setup MCP
4. Log your first decision — connected AI tools can recover the reasoning next session

---

14-Day Pro Trial

Every install gets full Pro access for 14 days — unlimited decisions, semantic search, knowledge graph, and all features. No credit card required.

After the trial, your decisions are preserved (read-only). Upgrade to Pro to continue logging.

---

Pricing

	Trial (14 days)	Pro
Decisions	Unlimited	Unlimited
MCP Integration	All tools	All tools
Semantic Search	Full	Full
Code Intelligence	Full	Full
Knowledge Graph	Full	Full

Pro Plans:

• Monthly: $9/month
• Annual: $89/year (save 18%)
• Lifetime: $199 one-time

Upgrade to Pro

---

Works With

• Claude Code
• Cursor
• Cline / Roo
• GitHub Copilot
• Google Gemini
• Any MCP-compatible AI tool

---

What's New

v2.16.7 — April 2026

• Fix: Pricing consistency — all CLI surfaces now show unified pricing ($9/mo · $89/yr · $199 lifetime). Removed non-existent $699 tier.
• Fix: Semantic search on free tier — license status now correctly shows semantic search as available (keyword fallback always works).
• Fix: TUI error message — tui/dashboard now suggest continuity serve instead of useless "reinstall the extension".
• Tag extraction expanded — +17 patterns: GitHub Actions, GitLab CI, Jenkins, CircleCI, Datadog, Grafana, Prometheus, Sentry, New Relic, CSS, Sass, ESLint, Prettier, deployment, testing, monitoring, observability.

v2.16.5 — April 2026

• Fix: graph and visualize commands — no longer crash on decisions with relationship objects (was "d.relationships is not iterable").
• Fix: scan respects free tier limit — scan results are shown but not logged when at the decision cap.
• Fix: CLI launcher auto-update — extension now rewrites the CLI launcher on every activation so version upgrades don't break the continuity command.

v2.16.2 — April 2026

• Fix: degraded-mode decision writes — when MCP is unavailable, decision logging now writes consistently to the local fallback path instead of silently dropping.
• Fix: CLI runtime hardening — removed vulnerable dependencies, hardened CLI startup.

v2.16.0 — April 2026

• Unified trial & license — extension and CLI now share the same trial countdown and license activation via ~/.continuity/trial.json and ~/.continuity/license-cache.json. Activating a license in either surface unlocks the other. 14-day active trial with 3-decision gate, matching the extension.
• continuity generate — new command produces CLAUDE.md, AGENTS.md, GEMINI.md, .cursorrules, and .github/copilot-instructions.md from project context. Supports --dry-run.
• continuity hooks install — install git pre-commit and post-commit hooks that auto-log decisions on every commit. Also hooks uninstall and hooks status.
• continuity init seeds from git history — first-run init now creates up to 10 draft decisions from recent commits so the sidebar isn't empty.
• continuity mcp setup auto-detects AI tools — scans for Claude Code, Cursor, Cline, Roo Code, Copilot, Gemini CLI, and Codex, then writes MCP configs automatically.
• MCP workspace targeting fix — global configs no longer get stuck pointing at the wrong project. Server detects workspace from cwd first; extension re-repairs on window focus and folder change.

v2.15.2 — April 2026

• CLI QA audit fixes — debug log noise removed from status and sync, graph/visualize commands rewritten (no longer crash), timeline dates use correct timezone, config supports get/set/list/reset subcommands, search --plain for non-interactive output, quick shows templates before paywall, relationships exits gracefully on cancel.
• Security: simple-git updated — 3.28.0 → 3.33.0 (critical RCE fix).

v2.15.0 — April 2026

• Fix: CLI missing dependency — rxjs was excluded from the VSIX but is required by inquirer at runtime. CLI no longer crashes on startup.

v2.14.9 — April 2026

• Fix: MCP workspace targeting — global MCP configs (Cursor, Claude Code, Codex, Gemini) no longer get stuck pointing at the wrong project. The MCP server now detects workspace from the client's working directory first, and the extension re-repairs configs when you switch windows or workspace folders.
• Fix: Onboarding analytics — funnel events (MCP auto-configured, first run seeded, onboarding complete) were defined but never wired. Now tracked correctly, with a one-time catch-up for existing users.

v2.14.6 — April 2026

• Fix: Install CLI command restored — "Continuity: Install CLI" is back in the command palette. It now uses the bundled launcher from the VSIX, so the CLI is available without npm link.
• Fix: CLI runtime — @continuity/core is now included in the VSIX so the CLI resolves its imports correctly after installation.

v2.14.3 — April 2026

• Work-already-done detection — log_decision now compares the decision text against recent git commit messages using Jaccard similarity. If a commit already covers the work (70%+ match), the LLM sees a warning before logging. Zero token cost — pure local git + string matching.

v2.14.2 — April 2026

• CLI expansion — 6 new CLI commands: debt (decision debt tracking), dream (memory consolidation), governance (conflict detection), notes (session notes), quick-log (templates & project scanner), scan (project analysis). Run offline without MCP.
• Handoff improvements — auto-save complete state on extension deactivation and window blur, interaction counter for proactive saving, session notes and cross-LLM guide included in auto-sync.
• Enforcement — branch switch detection via VS Code git API, file renames/deletes logged as decision debt, merge commits included in decision coverage audit, search-before-log enforcement in MCP middleware.
• Fixes — Thenable .catch compatibility fix, merge commit hook fix for worktree paths, MCP auto-repair for stale configs on version upgrade, auto-cleanup of old extension versions on activation.

v2.14.0 — March 2026

• Trial overhaul: active days + 3-decision gate — Trial no longer starts on install. Users get unlimited time with all Pro features unlocked until they log their 3rd decision. Then the 14-day countdown begins — counting only days the user actually uses Continuity, not calendar days. A user who uses Continuity twice a week gets 7 weeks of trial instead of 2.
• Pre-trial state — new lifecycle: pre-trial (all features, no countdown) → 3 decisions → active trial (14 active days) → expired → Pro.
• Welcome messaging — "Log 3 decisions to start your 14-day trial" on first install. "Your trial has started!" on 3rd decision.

v2.13.99 — March 2026

• Fix: Analytics tracking for all decision sources — DecisionCRUD.log() now fires decision_logged analytics event regardless of source (sidebar, MCP, git hooks, file fallback). Previously only MCP-originated logs appeared in the dashboard.

v2.13.97 — March 2026

• Set API Key button — Settings tab now has a clickable "Set API Key" button instead of text-only instructions. Uses VS Code SecretStorage (encrypted, never plain text).
• Cross-repo compatibility — new test suite and fixture files for validating Continuity across different repository shapes.

v2.13.95 — March 2026

• Removed MCP disconnect alerts — MCPWatchdog service removed. No more "MCP Down" status bar alerts or warning popups. MCP reconnection is handled by the AI client, not the extension. Reduces user anxiety from normal MCP lifecycle events.
• Cross-client improvements — WorkingMemoryService, HandoffArtifacts, AIInjector, and ProjectInstructionsGenerator updated for cross-client compatibility (Claude Code, Codex, Cursor, Copilot, Gemini).
• Documentation refresh — MCP server docs, integration guides, and getting-started docs updated.

v2.13.93 — March 2026

• Search-before-log enforcement — MCP middleware now tracks whether search_decisions or get_quick_context was called in the session. If log_decision is called without a prior search, a warning is injected into the response. Enforced at the tool pipeline level for all AI clients.

v2.13.91 — March 2026

• Cross-client MCP compatibility — WorkingMemoryService for cross-client memory management. FirstCallInjector improvements. Context recovery works across Claude Code, Codex, Cursor, Copilot, and Gemini.
• Dream Engine tuned — duplicate threshold 0.75→0.6, contradiction resolution 0.7→0.4, stale pruning 80→50. Dream now actually resolves contradictions and outdates stale decisions instead of finding issues and doing nothing.
• VSIX size halved — dream GIF compressed 24.9→7.4MB, VSIX 33.67→16.19MB.
• Broken test cleanup — removed 6 agent-built test files with wrong method signatures. Test failures 15→9 (remaining are pre-existing).
• Cross-client fallback replay tests — new test suite for MCP fallback behavior across LLM clients.

v2.13.87 — March 2026

• MCP hardening: deterministic, crash-safe, explicit — major reliability release focused on MCP correctness and durability.
• Read/write lock separation — multiple concurrent MCP reads, exclusive writes. Eliminates timeouts when AI assistants call parallel tools.
• Runtime parameter validation — every MCP tool call is validated against its schema before execution. Invalid params return clear error messages.
• Semantic governance conflicts — check_governance now uses TF-IDF cosine similarity with negation detection, not just keyword matching. Catches "disable HTTPS" conflicting with "use TLS everywhere."
• Jailbreak pattern hardening — 10 new patterns detect prompt injection, role reassignment, DAN mode, and bypass attempts.
• Decision schema validation — validates every decision on load and write. Malformed entries logged and skipped, never crash.
• Storage diagnostics — explicit workspace, decision counts, lock state, and config logged at MCP startup. getDiagnostics() API for programmatic health checks.
• Schema-handler consistency — startup check verifies every defined MCP tool has a matching handler. Dream tool registration gap fixed.

v2.13.85 — March 2026

• Fix: Governance lock self-locking — danger score now decays (10 points/min), manual reset command added, testing mode setting disables accumulation. Previously, security testing permanently locked all MCP tools with no recovery.
• Free tier works via MCP — expired trial users can now search/read decisions through MCP (was completely blocked). Graduated degradation instead of cliff.
• Security scanning improvements — scan_content now detects Base64-encoded payloads, URL-encoded attacks, nested HTML/script in comments, and event handler injection.
• Data integrity fixes — load_context_section("session_notes") path aligned, active/deleted decision dedup on load, stale file lock auto-cleanup with PID validation.
• Command palette cleanup — 13 agent commands and 5 team commands hidden via when clauses until enabled. Graph/timeline hidden until decisions exist.
• Draft decision promotion — auto-scanned decisions now created as active instead of draft (factual detections don't need review).
• 1,008 lines of new tests — governance lock, licensing, security scanning, data integrity.

v2.13.83 — March 2026

• Fix: Delete button — delete confirmation was silently failing in sidebar webview (modal dialog issue). Now uses non-modal notification that works reliably.
• Robust node path resolution — getNodeCommand() now checks well-known paths directly (/opt/homebrew/bin/node, ~/.nvm/*, ~/.volta/*) and auto-repair upgrades bare "node" to absolute path on every activation.
• Honest token validation test — rewritten to measure real data (actual CLAUDE.md + MCP responses vs manual file pasting) with current 2026 pricing for 5 models. Conservative >80% claim.

v2.13.80 — March 2026

• Fix: MCP "spawn node ENOENT" — MCP setup now resolves the absolute path to Node.js, fixing connection failures for nvm/fnm/volta users where VS Code's extension host doesn't inherit the shell PATH.
• Dream cooldown timer — 30-minute cooldown after dreaming prevents repeated clicks that return zero results. Per-workspace via localStorage. Early unlock: log 10+ new decisions to re-enable Dream before the timer expires.
• VSIX size reduction — excluded bundled CLI (42MB → 34MB, 5,152 → 379 files).
• Command palette cleanup — removed 9 unused/redundant commands (59 remain).

v2.13.78 — March 2026

• Initial Project Scanner — on first activation, Continuity scans your project's filesystem and auto-creates draft decisions for your tech stack, key dependencies, database, testing framework, deployment platform, API style, and project structure. No AI required — pure filesystem analysis in <1 second.
• Extended decision logging API — DecisionLogOptions now supports optional tags, status, and metadata fields for programmatic decision creation.
• Dream Engine fix — false-positive contradiction detection eliminated (was flagging 56 false contradictions, now requires shared conflict keywords between both decisions).

v2.13.76 — March 2026

• MCP response optimization — trimmed tool response payloads by ~60%. Removed verbose coaching text, ASCII art injections, and first-call-only workflow reminders. Reduces Claude Code context pressure.
• Decision caching — in-memory cache with file watching eliminates redundant 3.4MB JSON parses per tool call. Cache invalidates automatically on writes.
• Decision lifecycle: outdate instead of archive — old decisions are now marked status: outdated instead of being moved to a separate file. All decisions stay in one file, fully searchable, with the outdated counter showing correct counts in the sidebar.
• Dream Engine improvements — skips outdated decisions during consolidation. Fixed false-positive contradiction detection (was flagging 56 false contradictions, now requires shared conflict keywords between both decisions). Phase 4 marks stale candidates as outdated.
• Standardized MCP configs — all projects now point to the installed extension path with correct WORKSPACE_ROOT.

v2.13.73 — March 2026

• Zero-click MCP setup — Continuity now auto-detects your AI tools (Claude Code, Cursor, Copilot, Cline, Roo, Gemini) and configures MCP automatically on first activation when supported. No wizard, no manual steps for the supported clients.
• First-run decision seeding — on first launch, the sidebar is pre-populated with decisions from your git history so it's never empty.
• Welcome screen — first-launch experience with value props and one-click setup CTA.
• MCP setup banner — persistent "Connect your AI assistant" banner in the sidebar until MCP is connected, with 7-day dismiss.
• Quick Setup — one-click MCP configuration via command palette (Continuity: Quick Setup). Auto-selects detected tools.
• Onboarding analytics — 12 new activation funnel events to measure the onboarding improvements.
• Test coverage — 1,366 lines of tests across 5 new test suites for all onboarding services.

v2.13.71 — March 2026

• Dream Engine — 4-phase memory consolidation system that gathers signals, detects patterns, scores staleness, and consolidates decision memory. Animated dream overlay with Hackerware panda mascot.
• Knowledge Graph overhaul — node cap raised from 400 to 1500 (all decisions visible), tag filter now truly removes non-matching nodes instead of dimming, DAG layout modes re-render correctly, inline edit panel for clicking and editing decisions directly on the graph.
• MCP stability improvements — diagnosed and fixed spawn storm caused by 4 competing MCP consumers, softened process lock (no more SIGKILL), synchronous exit logging, 30s heartbeat keepalive.
• Decision enforcement — pre-commit audit blocks when too many unlogged commits, post-commit auto-logs stub decisions, fixed hook path mismatch between pre-commit and post-commit.
• Command wiring — registered viewDecisionDetail command (was called but never existed), fixed openDecision stub in Code Intelligence.
• Panda mascot — replaces Continuity logo in empty filter state with Hackerware panda.
• Codebase pruning — removed 5 unused native dependencies, 2 dead services, deprecated settings, test commands. ~625MB of dev artifacts cleaned.

v2.13.65 — March 2026

• Knowledge Graph rendering fix — re-applied Array.isArray guards for non-array tags that crashed the webview silently, rendering 0 nodes. Added NaN coordinate guard and onEngineStop zoomToFit fallback for reliable rendering.
• Knowledge Graph seed algorithm — graph now seeds connected decisions (those with relationships) first, then fills with recent decisions, so the graph is populated even when recent decisions lack relationships.
• MCP watchdog — new self-healing reconnection service that detects MCP disconnections and automatically restarts the server.
• Health dashboard feedback loop — MCP instrumentation and test coverage improvements.

v2.13.58 — March 2026

• MCP server health monitoring — new get_mcp_health tool reports current uptime, session history, restart count, exit reason breakdown, and avg session duration. Session data stored in .continuity/mcp-sessions.jsonl.
• MCP server logs fixed — get_mcp_logs tool was defined but had no handler since Nov 2025. Now returns recent log buffer entries with level filtering.
• Decision debt dismissal fix — dismissed debts no longer re-activate on every file scan. Dismissals now persist correctly.
• MCP config fix for Claude Code — removed type: "stdio" from Claude Code's .mcp.json (its schema rejects it; stdio is implicit). Other tools still get it.

v2.13.55 — March 2026

• MCP config schema fix — Added missing type: "stdio" to all 6 MCP config generators (Claude Code, Claude CLI, Cursor, Cline/Roo, Gemini CLI, Antigravity IDE). Fixes "Does not adhere to MCP server configuration schema" validation errors that caused connection failures.

v2.13.54 — March 2026

• Streamable HTTP dual-transport — MCP server now supports HTTP transport alongside stdio for improved resilience. Server survives client disconnects and supports multiple simultaneous sessions. Activate via CONTINUITY_TRANSPORT=http env var.
• Decision logging analytics — track decision_logged events in DecisionCRUD for engagement measurement.

v2.13.52 — March 2026

• Decision logging analytics — track decision_logged events with running decision count for engagement measurement.
• Analytics security — migrated analytics endpoint from HTTP to HTTPS via DigitalOcean app URL for improved security and reliability.

v2.13.48 — March 2026

• Knowledge Graph — Obsidian-style force-directed graph visualization of decision relationships. New continuity.openGraph command with keybinding.
• MCP resilience — fallback CLI instructions added to CLAUDE.md so decision logging continues even when MCP disconnects mid-session. PostToolUse hook detects disconnects automatically.
• MCP handshake fix — deferred startup coverage check no longer blocks the MCP server handshake.
• Repo cleanup — removed 44 unnecessary files from the repository root.

v2.13.45 — March 2026

• MCP server stability fix — eliminated SIGPIPE death spiral that caused server to become a zombie process. Signal handlers no longer write to broken pipes.
• Delete decision button fixed — the Delete button in the sidebar now works (was silently blocked by VS Code webview confirm() restriction).
• stdout safety hardening — esbuild banner ensures console.log is redirected to stderr before any dependency code loads, preventing JSON-RPC protocol corruption.

v2.13.42 — March 2026

• Onboarding analytics funnel — 20 new event types track every step from activation to first decision, MCP setup, and beyond
• Better onboarding UX — no-workspace state now has an "Open Folder" button; generic projects get a useful first-decision suggestion; MCP setup banner persists in sidebar until configured
• Improved aha moment — post-first-decision simulation shows a mock AI conversation instead of raw context dump
• Re-engagement nudges — trial users inactive for 3+ days get a gentle reminder to log decisions
• Weekly value summary — after 5+ decisions and 7+ days of use, see how many times your AI recalled decisions

v2.13.40 — March 2026

• Trial fix — new users no longer incorrectly show as expired; trial auto-initializes on first activation
• Analytics reliability — events flush immediately for key actions, race condition with license check fixed
• Sidebar performance — removed log spam and unnecessary file-change processing when sidebar is hidden

v2.13.30 — March 2026

• 14-day Pro trial — full Pro access on every install, no credit card required
• Analytics service — anonymous usage telemetry to improve the product
• Sidebar search fix — hybrid, semantic, and keyword search modes now route correctly
• MCP latency optimizations — batched git commands, pre-filtered dedup, non-blocking embeddings
• Tool handler hardening — safe JSON parsing, governance tags parameter, 8 error paths fixed
• MCP tool call instrumentation — duration and success tracking per tool call
• Test coverage — 50% → 70% with 1,200+ new tests across 15 services
• In-app changelog — see what's new directly in the extension
• LicenseManager updates — improved trial flow and license validation

v2.13.22 — March 2026

• Inline decision editing — edit decisions directly in the sidebar without opening a modal
• Delete button on decision cards for quick cleanup
• Decision status consistency fix — no more phantom disk writes on reload
• Single-instance MCP lock file prevents server process pileup
• API key storage migrated to VS Code SecretStorage (security hardening)
• MCP latency reduction — decision caching and capped impact traversal
• Marketplace review prompt after 5th logged decision
• 16 test suite failures fixed

v2.13.21 — March 2026

• Health dashboard — monitor extension status, MCP server health, and decision storage at a glance
• Added /save-all skill for quick commit + decision logging workflow
• Updated AI instruction docs (CLAUDE.md, AGENTS.md, GEMINI.md, Copilot)
• Removed deprecated dmux-hooks directory

v2.13.19 — March 2026

• Live sidebar updates — decisions, stats, and milestone bar refresh automatically when decisions are logged via MCP
• Confetti celebration animation on milestone achievements (1, 5, 10, 25, 100, 500, 1000+)
• Replaced text icons with Continuity logo across all sidebar tabs
• Security hardening — patched command injection, path traversal, and parameter logging vulnerabilities
• Streamlined MCP toolset — removed 18 bloat/stub tools (59 down to 41) for faster, cleaner AI integration
• Team collaboration tab (preview) — invite codes, member list, sync status
• Onboarding flow — stack detection, pre-filled first decision, aha moment simulation

v2.13.17 — March 2026

• Corrected MCP tool profile counts in Settings UI and notifications
• Extended milestone progress bar to support 10,000+ decisions
• Sidebar stability fixes — Settings tab, message handler reliability

v2.13.4 — March 2026

• Removed render-blocking resources from sidebar
• Added "Open a Folder" prompt when no workspace is open

---

Support

• Issues: GitHub
• Email: support@hackerware.com