Development Roadmap

The planned evolution of Memo across upcoming releases.

v3.1.0 — Foundation Expansion ✅

Released — 10 major features and 61 bug fixes.

• WhatsApp integration via Baileys bridge
• External LLM providers (OpenAI, Anthropic, Google, Ollama)
• Autonomous agent system with tool pipeline
• Orchestra mode with conductor pattern
• Mobile app (iOS + Android) initial release
• Encrypted backup and restore
• 61 bug fixes (13 critical, 15 high, 13 medium, 20 low)

v3.1.x–v3.2.x — Terminal CLI ✅

Released — built from scratch, then hardened and redesigned.

• Terminal CLI (Claude Code–style chat REPL), autonomous task loop
• CLI runs as its own process, independent of the desktop app
• Two rounds of reliability fixes and a full welcome-panel redesign
• File edit tool, git-adjacent agent tooling, improved error recovery in agent pipelines

v3.3.3 — Routines, Proactive Learning & Memo Swarm ✅

Released July 23, 2026 — see Release Notes for the full write-up.

• Routines (scheduled automations, desktop + mobile)
• Proactive Learning & Ambient Nudges, Self-Insight (/insight)
• Notification system for proactive suggestions (desktop suggestion banner)
• Web search tool toggle now in Chat's top bar
• Model Store/Discover overhaul: hardware-aware setup, concurrent downloads, real capability detection
• Two more providers: OpenCode Zen, OpenCode Go (9 total)
• Developer API Gateway (Anthropic-compatible local endpoint)
• Memo Swarm (beta) — pooled-compute for oversized local models
• Usage Stats tab, Import Memory From Another AI
• Security: Remote Access token enforcement, symlink sandbox-escape fix, Claude empty-model-field fix, dependency patches

v3.3.4 — Reliability, Live Mode & Tailscale Out of Beta ✅

Released August 5, 2026 — see Release Notes for the full write-up.

• Backend-wide panic recovery — a background-task crash no longer takes the whole app down
• Fixed a 4-5x local generation slowdown caused by memory/RAG being on (embedding/chat model VRAM contention)
• Live Mode (beta) — hands-free voice chat, local Piper TTS by default, offline voice picker, one-directional barge-in
• Claude Code / Codex CLI as chat providers (beta) — per-chat background coding agents
• Tailscale remote access graduates out of Beta — one-click login, Funnel default-on, auto-reconnect
• Settings reorganized into a searchable rail instead of ~20 flat tabs
• @ file-mention in chat, quick model/provider switcher pill
• Fixed: agent mode failing on short messages with small-context local models, "Delete All Data" on Windows, STT from an installed CLI, web search firing on every message, missing msvcp140.dll on clean Windows installs

v3.5.5 — Self-Hosting ✅

Released August 17, 2026 — see Release Notes for the full write-up.

• Full 4-mode auth system (none/token/password/token+password), argon2id hashing, per-device tokens
• Multi-account support with admin/user roles, a universal auth gate across every screen
• Real Flutter web UI replacing the old hand-rolled headless/CasaOS client
• Docker/CasaOS images (multi-arch GHCR) and dedicated server-only installers (get-memo-server.sh)
• Complete CLI management toolkit — memo config/remote/service/provider/agent/model over SSH
• Full mobile-responsive pass across every screen
• Orchestra Mode + Agent Mode now genuinely share real tool access
• Fixed: Task Loop silently doing nothing on start, Orchestra's utility calls forced through the wrong pipeline, agent mode skipped by non-streaming chat
• Mood off by default; CORS bypass and a net/url dependency vulnerability closed

v3.9.0 — Talk to Memo ✅

Released August 22, 2026 — see Release Notes for the full write-up.

• WhatsApp self-chat assistant — message yourself, Memo replies with full chat/memory/agent capability
• Telegram integration, brand new — connect a bot token, first-sender owner lock, same self-chat assistant pattern
• Routines now creatable, listable, and cancellable straight from a WhatsApp or Telegram conversation — delivery hard-locked to the surface you're talking from
• Self-hosted multi-account gets 7 independent per-capability permissions (Models, Memory, Agent, Calendar, WhatsApp, Telegram, Routines), enforced server-side
• Kilo Code joins the provider list (10 total) with a live free/paid model browser; the same treatment extended to OpenCode Zen
• Web search rebuilt around real per-message tool-calling instead of injecting the raw message every turn
• Reasoning-effort control rebuilt on live per-model capability discovery (Claude/Gemini/Ollama/OpenRouter) instead of static tables
• Developer API Gateway redesigned — new OpenAI-compatible endpoint (/v1/chat/completions), one-click Claude Code CLI connect, configurable system prompt
• System tray icon, Dream's own configurable schedule, a Stats token-spend-by-category breakdown, a visible "N memories used" badge
• Mobile navigation redesign — hamburger drawer replaces the desktop NavRail below 600px; routine delivery to the phone app removed (superseded by WhatsApp/Telegram)
• Security: a Cloudflare Tunnel auth-bypass closed (loopback trust didn't account for a local proxy forwarding external traffic), the setup wizard no longer re-appears per-origin, plus three previously-open High-severity issues resolved
• Fixed: self-chat's agent tools were never actually reachable — the most consequential bug of this stretch, now fixed unconditionally, the same fix applied to Routines' own agent access

v3.6.0 — Extensibility

Planned

• Plugin system with sandboxed JavaScript/WASM runtime (skill tools already run real commands as of v3.3.3 — this extends that to a full marketplace)
• Plugin marketplace for community contributions
• Custom tool builder UI (no-code tool creation)
• Webhook triggers for automation
• Memo Swarm out of beta, including macOS support
• Conversation branching and forking

v4.0.0 — Self-Improving Intelligence

Planned

• Knowledge graph construction from memories
• Self-improving memory consolidation and pruning
• Deeper contextual awareness across sessions and devices
• Long-term goal tracking and progress monitoring
• Shared conversation links (read-only, expiring)
• Offline voice transcription improvements, full mobile feature parity