Openhuman changelog: what's new each month
Every stable Openhuman release summarised by month: the highlights, new features, improvements, fixes and anything you need to act on. 8 months covered; the current month updates daily.
October 2026
2 releases: 0.64.14 → 0.64.15
October brought major usability improvements centered on clearer file and session management, a significant evolution to memory handling, and real-time voice agent capabilities. The month also focused on hardening imports, migrations, and reliability across the platform.
- In-app model downloads have been removed
- Ollama lifecycle management has been removed; local inference now requires user-run endpoints
Highlights
- Real-time voice agents are now available
- Memory system received a major evolution
- Imports and migrations are significantly more reliable with clearer states and better failure visibility
- Simplified local AI support by removing in-app model downloads and Ollama management
- Improved stability of key modules including search and memory
New
- Real-time voice agents
- Retry controls for import and migration failures
- Better keyring support for headless deployments
- Signature verification for macOS module bundles
Improved
- File and session clarity in the interface
- Import and migration batching and retry logic
- Failure visibility during imports and migrations
- Resume reliability for interrupted operations
- Web research budget handling
- Credential scrubber behavior to avoid unwanted code redaction
- HTTP status code handling
- Orchestration behaviors
Fixed
- Web research budget issues
- Credential scrubber incorrectly redacting code
- HTTP status code handling errors
- Stability when loading search and memory modules
- Orchestration behavioral issues
September 2026
5 releases: 0.64.0 → 0.64.10
September brought a focus on stability and performance across all platforms. The 0.64.x series delivered faster turn startup, more reliable offline operation on Windows, and improvements to the core experience end-to-end.
Highlights
- Faster first turn startup and reduced processing loops under real workloads.
- Improved Windows desktop experience with better offline reliability.
- Clearer billing and usage visibility surfaces.
- Sturdier integrations and more reliable recovery behavior.
- Internal modularity improvements enabling faster future development.
Improved
- Chat feedback now clearer and more responsive.
- Windows offline operation smoother and more polished.
- Core experience tightened from end to end with performance refinements.
- Integration robustness increased across the platform.
Fixed
- Reduced unnecessary loops in agent processing.
- Better recovery behavior under real-world workloads.
- Windows desktop stability improvements.
August 2026
3 releases: 0.63.11 → 0.63.7
August brought improvements to voice agent integration, embedding reliability, and platform stability. The month focused on fixing critical issues with TTS providers, embedding failures, and session management.
Highlights
- ElevenLabs Voice Agents support with core RPC, realtime sessions, and local-agent relay handling.
- Fixed signed-in user embedding failures and unstuck the parked queue.
- Resolved Piper voice ID leaking into non-Piper TTS providers.
- Improved embedding budget handling with warnings before silent memory stops.
- Better actionable error classification for Ollama unavailability.
New
- ElevenLabs Voice Agents with realtime session and relay handler support.
- Relay identity token transmission to backend.
Improved
- Embedding budget now warns before silently stopping memory.
- Ollama unavailability errors are now classified more actionably.
- Agentbox configuration handling with better base URL normalization.
- Conversations UX improved to prevent thread-goal footer auto-scrolling.
- Coding-session backlog drained across bounded passes.
- Windows keyring build stability enhanced.
Fixed
- Piper voice ID no longer leaks into non-Piper TTS providers.
- Execute bit on extracted Piper binaries now properly repaired.
- Signed-in user embeddings failing issue resolved.
- Parked embedding queue now unstuck.
- U16 sample format now handled in fallback capture path.
July 2026
5 releases: 0.61.2 → 0.63.9
July brought improved resilience and robustness to Openhuman's core agent execution. The month focused on making tool calls and API interactions more reliable, especially when dealing with imperfect model output.
Highlights
- Tool calls now recover gracefully from garbled output produced by smaller models.
- The /announcements/latest endpoint returns null on 404 instead of erroring.
- Composio task execution was tightened for faster and more reliable agent performance.
- Release pipeline improvements kept builds stable and deployments fast throughout the month.
Improved
- Tool-call robustness improved with recovery from pipe-garbled <toolcall> tags.
- Announcements API resilience improved to handle missing resources gracefully.
- Composio task execution made more reliable.
- Release pipeline and shipping process streamlined for smoother deployments.
Fixed
- /announcements/latest no longer errors on 404 responses.
- Tool calls now recover from malformed output from small language models.
June 2026
12 releases: 0.57.5 → 0.58.7
June brought major orchestration and multi-agent capabilities to Openhuman, alongside stability improvements and expanded provider support. The month focused on agent coordination, memory efficiency, and desktop reliability.
Highlights
- Multi-agent orchestration with personalities, scoped memory, and delegation control
- Model council feature enabling parallel multi-model deliberation with chair synthesis
- Agent workflows with phase-keyed task lifecycle playbooks
- Live streaming subagent view with reopenable sub-threads and persistent runs
- Expanded provider ecosystem including OMLX, ModelScope, and Kimi with streamlined setup
New
- Multi-agent personalities with scoped memory and delegation
- Orchestration control plane for managing agent coordination
- Multi-model deliberation with parallel members and chair synthesis
- Agent workflows with phase-keyed task lifecycle playbooks
- Live streaming subagent view with reopenable sub-threads
- Per-integration health strip in memory tree
- Vault file-system watcher implementation
- ArtifactPending event publishing and socket bridge
Improved
- Iteration turn cap now applies only to short-running agents
- Memory sync with bounded-concurrency GitHub item reads for better performance
- Multimodal support enhanced with FILE markers and text extraction
- Agent action sandbox now separated from internal workspace state
- Chat layout improved on short windows
- Compatible-provider timeouts made configurable
- Approval prompt visuals improved for better readability and contrast
- Agent auto-complete behavior strengthened with improved goal completion directive
Fixed
- Orchestrator integration context now refreshes on mid-session OAuth
- Inference errors now surface actionably when Managed route fails with no credits
- Strict PII rejection now excludes bare-phone patterns to reduce false positives
- Intelligence tab hang on shm-incompatible filesystems fixed by degrading WAL journal mode
- Telegram approval gate now wired end-to-end for Telegram turns
- AppImage crashes caused by glibc mismatch resolved by ensuring bundled ld-linux replacement from host system
May 2026
9 releases: 0.53.16 → 0.56.0
May brought major improvements to memory management, mascot features, and observability across Openhuman. The month culminated in activating the chat memory pipeline and embedding storage system for better conversation recall.
Highlights
- Chat memory pipeline now active with automatic recaps and cross-thread recall.
- Floating desktop mascot on macOS via native integration, plus new animations.
- Transcript-to-memory ingestion pipeline to automatically capture conversation content.
- Per-model embedding storage system live with migration and backfill complete.
- Improved error reporting with Sentry filtering to reduce noise from transient issues.
New
- Floating desktop mascot for macOS using native NSPanel and WKWebView.
- Seed phrase onboarding and multi-chain wallet setup.
- Microphone input device selector for voice composer.
- Memory tree UI with cloud-default LLM, queue priority, and entity filtering.
- Transcript-to-memory ingestion pipeline.
- Coding-harness baseline primitives for tools.
- Centralized error reporting with Sentry tags.
- Profile mirroring from connected accounts into PROFILE.md.
Improved
- Chat memory pipeline with Archivist wiring, LLM recaps, and segment-granular tree.
- User message bubbles now size to fit text content.
- Turn state persists and conversation history restores on cold-boot.
- Faster webview account cold opens and clearer UI.
- Local AI off by default with scoped usage flags and faster auto-updates.
- Billing interface drops rate-limit UI and surfaces cycle usage insights.
- Thread selection preserved on cold-boot identity hydration.
- Version, shutdown, and update RPCs with mid-thread integration refresh.
Fixed
- Cross-user state and CEF cookie leak in authentication.
- Assistant replies no longer duplicate on multi-segment turns.
- Memory tab hidden from bottom navigation bar.
- Progress UI re-syncs on turn boundary.
- Main window hides properly on Windows close.
- Misconfigured API base URL can no longer corrupt routes.
April 2026
26 releases: 0.50.3 → 0.53.4
April brought major improvements to onboarding flows, agent capabilities, and channel integrations. The month focused on stabilizing core features, expanding local AI support, and consolidating the desktop experience.
- Removed Screen & Accessibility Permissions step from onboarding.
- Disabled email login in favor of OAuth-only authentication.
- LocalAI locked to 2-4 GB model tier by default until explicit opt-in.
- Moved Tools step from onboarding to Settings.
Highlights
- Redesigned onboarding with conversational stack, skill integration (Gmail, Calendar, Drive, Notion), and streamlined permission handling.
- Introduced agent-scoped tool selection, per-agent profile injection, and smart routing for better task execution.
- Added Telegram and Discord channel support with threading, live listeners, and dynamic message handling.
- Implemented local AI model tier selection by device capability with cloud fallback for low-memory devices.
- Consolidated overlay into desktop app with webview notifications and improved voice dictation across platforms.
New
- Parallel search provider for WebSearchTool.
- Subconscious loop for local-model background awareness via heartbeat.
- Memory RPC surface with entity types in recall/query API.
- Developer webhook debug panel and endpoint updates.
- Referral system with link-based claims and $5 rewards.
- Screen intelligence OCR-only mode without vision model.
- Hierarchical summary tree module with CLI.
- Gmail integration and LinkedIn profile enrichment in onboarding.
Improved
- Reduced agent loop hallucination and improved tool call reliability.
- Enhanced voice dictation with hallucination filtering and GPU acceleration support.
- Standardized onboarding UI with consistent Next/Continue buttons and better error notifications.
- Light theme redesign across onboarding, navigation, and page layouts.
- Improved autocomplete observability, reliability, and sanitization with debug logging.
- Billing and subscription UI redesign with normalized API responses.
- Skill authentication unified with start({validate}) handshake.
- Increased Rust unit test coverage to ≥80% across critical modules.
Fixed
- Fixed onboarding state persistence and overlay reappearance on logout.
- Resolved screen intelligence screenshot capture and settings flow stability.
- Fixed Windows MSVC build with whisper-rs-sys patch.
- Corrected LocalAI download failure notification and retry flow.
- Fixed Ollama workspace install, download retries, and progress reporting.
- Resolved Telegram threading, live listeners, and webhook cleanup issues.
March 2026
1 release: 0.49.32
March brought refinements to testing infrastructure, updates to landing page presentation, and stabilization work on core daemon services.
Highlights
- Improved daemon service lifecycle stability and aligned Agent Status UI behavior
- Updated landing page with revamped design
- Enhanced testing scripts and dependency management
New
- Gitbooks integration support
Improved
- Daemon service lifecycle now more stable
- Agent Status UI better aligned with service state
- Testing scripts refactored for improved maintainability
- Dependencies updated to current versions
Fixed
- Monday-related patches applied
- Wednesday-related patches applied
Summaries are written automatically from the official release notes (full changelog ↗); check the original notes before relying on a detail. Openhuman: pricing, features and alternatives · All changelogs