Changelog
Release history, commit by commit.
Twenty-nine milestones mapped to liminal-ai pushes. We are on v0.1.2 (beta) today. RC and install GA are still ahead on the roadmap.
Liminal 0.1.2 — multi-account mail, inbox watcher, file attachments
Liminal AI v0.1.2: scan all connected mailboxes with account-aware replies, automated inbox watcher, email privacy redaction, drag-and-drop any file into the composer, integrations hub polish, and managed inference Settings fixes.
Liminal 0.1.1 — desktop auto-update
Liminal AI v0.1.1: portable desktop installs check GitHub Releases on launch, hot-swap harness updates in place, and schedule full app updates with a one-time restart.
Liminal 0.1.0 — first public preview (beta)
Liminal AI v0.1.0: first tagged public preview — CI on main, eval sandbox framework, inference latency mode, desktop managed inference catalog, and sidecar multi-chat registry.
Liminal 0.0.26 — live compose dock, email preview, Azure ARM & Kimchi
Liminal AI v0.0.26: immediate desktop compose preview for file edits and email drafts, rendered HTML in the email dock, expanded Azure Resource Manager tools, Kimchi/Cast AI provider throttling, and OSC 133 shell integration for the agent terminal.
Liminal 0.0.25 — desktop polish, Sheets auto-fit & reliable integrations
Liminal AI v0.0.25: Flutter design system, Vireon hub and chat UI refresh, Google/Linear/Sheets tool-arg normalization, automatic Sheets column sizing, and a structural fix so research turns always finish with a chat reply.
Liminal 0.0.24 — hosted OAuth connectors & desktop integrations hub
Liminal AI v0.0.24: hosted OAuth bridge that connects Google Workspace, Microsoft 365, GitHub, Slack, Linear, Notion, and Xero to the desktop harness with no per-user app registration, plus a hosted token-refresh API, custom GitHub website login, and a desktop-first integrations hub.
Liminal 0.0.23 — Microsoft 365, email dock & lazy tools
Liminal AI v0.0.23: Microsoft 365 OAuth + Graph REST + MCP, Google Workspace REST hybrid, GitHub connector, lazy tool-family loading, enterprise email compose with live desktop preview, integrations hub, and orchestrator handoff.
Liminal 0.0.22 — Serper search, agentcard & browser dock
Liminal AI v0.0.22: Serper-backed web search with DuckDuckGo fallback, agentcard harness tools, Gmail compose guard, desktop browser dock, default persona on bootstrap skip, and liminal widget apps gated off until stable.
Liminal 0.0.21 — Plan progress UI, chat replay & desktop voice
Liminal AI v0.0.21: single in-place plan progress on web/TUI/desktop, session replay hydration, sidecar TTS/STT and Flutter dictation, persona UI v2 copy/layout/invariants, provider token scrub.
Liminal 0.0.20 — Vault knowledge graph & desktop provider presets
Liminal AI v0.0.20: per-entity vault dossiers (vault_ingest_entities, vault_recall), desktop Settings provider presets, and managed-inference BYOK fallback to free owl-alpha when credits are exhausted.
Liminal 0.0.19 — Desktop downloads for Windows, Mac & Linux
Liminal AI v0.0.19: pre-built desktop app on GitHub Releases for all three platforms, install-page download cards, and portable liminald bundling for shipped builds.
Liminal 0.0.18 — Desktop (Windows), routing recovery, reasoning-stall breaker
Liminal AI v0.0.18: Windows desktop alpha with sidecar protocol, OpenRouter provider-ignore recovery on long runs, reasoning-stall breaker, and managed-inference proxy fixes.
Liminal 0.0.17 — team shared memory & cloud sync APIs
Liminal AI v0.0.17: Team-tier org memory sync across machines, Pro cloud sync on the control plane, NotesFacade contracts in CE, and Settings visibility for team memory status.
Liminal 0.0.16 — managed inference, dynamic routing & Pro sign-in
Liminal AI v0.0.16: run without your own API key via Vireon-managed inference with included credits, price-sorted OpenRouter routing, Enterprise Edition auto-install on Pro login, dictation privacy defaults, and a security-hardening pass.
Liminal 0.0.15 — open-core licensing, Enterprise Edition & dynamic workflows
Liminal AI v0.0.15: open-core entitlements with Ed25519 license tiers, the Vireon control plane (Stripe + Supabase billing), a separate Enterprise Edition package, and dynamic multi-phase workflows.
Liminal 0.0.14 — memory curator, output effort & voice I/O
Liminal AI v0.0.14: LLM memory curation with soft-delete, output-effort dial, TTS/dictation, recall reranker, and prompt caching.
Liminal 0.0.13 — provider presets & settings completeness
OpenRouter model packs in web Settings, full harness settings catalog coverage, simpler turn-end pipeline, Android scaffold removed.
Liminal 0.0.12 — research ledger & FSL license
Research ledger and research_state tool, LLM identity memory (user:name), inline web chat switcher, recall_relevant schema fix, optional PASTE speculation.
v0.0.11 — user-global storage & multi-chat
User-global ~/.liminal storage, web multi-chat, audio transcription, OpenAPI and MCP connectors, memory federation, prompt cache.
v0.0.10 — execution state & dev CLI
Task Worlds removed; execution state and compensation wired, liminal dev CLI, recipe library v2, web_fetch and persona theme polish.
v0.0.9 — writes, reasoning, and browser
Streaming large-file writes with resume, reasoning budget and intent routing, Playwright browser + CAPTCHA, compensation ledger, and MIT license added to repo.
v0.0.8 — persona product pass and settings
Typed harness defaults, Obsidian vault discovery, web_fetch hardening, persona bootstrap and PersonaUiTheme, web Settings modal, orchestration expansion.
v0.0.7 — world context and session traces
Harness rule recall, runtime prefs, agenda and scheduler tools, auto-dream telemetry, session JSONL, and platform context gatherers.
v0.0.6 — research, MCP, and extensible tools
Unified write_file/edit_file surface, grep_file, source credibility tiers, dynamic tools, MCP client, and tool progress streaming.
v0.0.5 — Upgrade V (orchestration and memory bus)
Shared memory bus, spaced-repetition ranking, intra-round tool DAG, semantic context compression, and goal/branch harness tools.
v0.0.4 — platform tools and long-horizon runs
Streaming retry UX, vision sidecar, multi-market quotes, memory-first recall, optional self-heal lint, and document engine scaffolding.
v0.0.3 — hardening, recall, and lazy tools
Tool argument guard, vault path resolution, hybrid memory retrieval, safety judge, workspace-root grounding, and lazy tool family loading.
v0.0.2 — persona generation and harness wiring
LLM-generated personas, strength dial and modifiers, onChildCreated child tools, and early orchestration limits.
v0.0.1 — monorepo and first loop
Workspace scaffold, AgentHarness ReAct engine, initial tool catalog, Ink TUI, web SSE UI, eval harness, and Liminal rebrand.