Changelog

Release notes

Every release of ROW Recall, documented for clarity and transparency.

0.8.0 — 2026-08-15 · Memory Flywheel & Ask ROW

Ask ROW — natural-language questions answered from your own timeline, memories and agent sessions, fully local with no AI needed: "What changed today?", "Why was this file modified?", "What did the last agent do?", "What's the last known good state?", "What do we know about <topic>?". Structured answers with evidence links straight to the timeline.
Experiment memory — every rollback / undo / redo automatically records an "Experiment rejected" memory, so agents and humans stop re-trying approaches that already failed.
Memory lifecycle — verify, deprecate (with a reason) or archive any memory; deprecated and archived memories are excluded from context, so stale knowledge can't mislead.
Provenance deep-links — "View evidence" jumps from any memory to the exact timeline event it came from.
recall_ask MCP tool (21 tools total) + recall ask CLI — the same local Q&A engine from every surface.

Persistent memory layer — project, decision, failure and agent memories with provenance, confidence and context ranking (relevance / recency / confidence).
Workspace Intelligence (Day-1 card) — detects your stack from manifests and gives useful context within minutes of adding a workspace, no history required.
recall_memory MCP tool (20 tools total) — search development memory with a type filter; recall_context now carries top memories + the intelligence card.
Per-workspace timeline separation — browser-style workspace tabs, global dashboard totals, and AI sessions scoped to their own workspace.

recall_context() and recall_search — the killer MCP tools: one call returns a compact context package (workspace state, recent changes, AI sessions + files touched, last checkpoint, hot files) with a token budget. Minimal Context / Search / Action surface, 19 tools total.
A/B benchmark harness — measured claim machine: same task with vs. without recall_context, per-run JSON metrics and a summary table.
Cross-device recovery — account/device model, encrypted blob transport, Cloud Recovery UI, and a full new-PC restore flow. The backend stores only ciphertext.
AI session file attribution — every agent session shows the exact files it edited; restore to before any session with one click.
Local database fully encrypted (SQLCipher / AES-256), machine-bound licensing with encrypted delivery, plugin marketplace for Claude Code & Codex, @rowai/mcp-server on npm.

Full-database encryption — AES-256 (SQLCipher) at rest, with an OS secure-store key and a hardware-recovery copy that survives OS reinstalls.
AI session file attribution — every agent session shows the exact files it edited, with one-click "restore to before this session".
Parallel memory layer (one watcher per workspace), database disk hygiene, timeline noise reduction, and keyset pagination.
Plugin ecosystem — @rowai/mcp-server on npm and the Claude Code plugin marketplace.
Development release — no installers shipped; these features are included in the 0.6.0 build.

Production domains — rowrecall.com and brain.rowrecall.com, license server + admin dashboard, realtime stats.
Team accounts with roles (admin / support / accounts), support tickets, customer registration and CSV export.
Desktop first-run signup with OTP email verification and welcome Pro license by email.
Installer & auto-update pipeline with signed releases, website SEO + privacy/terms pages.

@rowai/mcp-server (17 tools) + plugins for Claude Code, OpenAI Codex, Cursor and Windsurf.
Free/Pro tiers (100-event cap / unlimited), multi-workspace support (up to 5), AI activity manager, team webhook, Stripe license server.
Timeline pagination and CPU-metric fixes.

Renamed to ReCall by Row AI Technologies; dark premium UI overhaul with replay mode.
Rollback, per-file restore, unified diff, move detection, branching, undo/redo engine, CLI and VS Code extension.
AGPL v3 with commercial licensing; WebSocket origin check; live system metrics.

Tauri v2 desktop app: SQLite WAL engine, kernel file watcher, WebSocket agent server, dual-pane UI.