Release v0.4.1

Fixes

Neo4j Plugin — Crash on Startup

• Fixed a panic in neo4rs 0.8.0 (connection.rs:287 — unwrap() on None) caused by plugin ABI probes firing real database queries before user credentials were injected via configure.
• Removed all pre-configure functional probes from the plugin loader. Liveness verification is now done post-configure via nil-input smoke tests with no external side effects.

OpenBao — Missing mount Config

• Added the mount field to OpenbaoSecretsConfig (defaults to "secret"). Users migrating from 0.3.0 with a non-standard KV mount path can now configure it without env-var workarounds.

MCP — Tokens Blocking Server Connections

• Fixed a bug where missing OAuth tokens in the secrets backend would permanently block MCP servers from connecting (pending-auth). Servers that don't require authentication now connect without a token.

Plugin-Level Debug Logs Leaking at Info Level

• The emit_log JSON-RPC handler now routes debug-level messages through the logging framework, respecting the configured log level. Previously, all plugin messages printed unconditionally.
• Moved Telegram inbound/outbound message content and Ollama API call details from info to debug level.

Backward Compatibility

Validation Layer — 0.3.0 Schema Check

• Added utils/validation_layer/src/backcompat/ to automatically verify that all 10 built-in plugin JSON schemas remain backward-compatible with the 0.3.0 Viper-based config. Detects renamed properties, changed types, and missing required fields at plugin validation time.

Neo4j Config — user / username Alias

• The Neo4j plugin now accepts both user (0.3.0 Viper key) and username (current schema key) at runtime, ensuring existing configs work without modification.

Schema Corrections

• Neo4j: Missing "required" field — added ["uri", "user", "password"].
• Slack: app_token was missing from required alongside bot_token.
• Secrets JSON: Added "required": ["path"].

Defaults

• A2A: Disabled by default (a2a.enabled: false), matching setDefaults().

Release v0.4.0

What's New

Native Plugin System Migration

• JSON-RPC Runtime: Completely replaced the internal adapter system with a robust native runtime using JSON-RPC. This provides superior isolation, security, and performance.
• Official SDK Support: Plugins are now built using official SDKs in Rust and Go (TinyGo), ensuring stability and better integration with third-party libraries.
• Dynamic Discovery & Validation: Improved the plugin discovery service with strict JSON schema validation and runtime state recovery.
• Built-in Plugin Refactor: Migrated all core adapters (Anthropic, OpenAI, Ollama, etc.) to the new native architecture.

Multimodal & Audio Support

• Audio Plugin Architecture: Introduced a dedicated audio processing layer in the plugin system.
• ElevenLabs Integration: Added official support for ElevenLabs TTS/STT as a high-quality fallback for voice interactions.

Deployment & Installation

• One-Click Local Installer: Added scripts/install.sh for instant setup on Linux, macOS, and Windows (WSL2), including automatic service configuration for systemd and launchd.
• Cloud Quick-Deploy: Updated infrastructure templates to support "One-Click Deploy" for Azure, Google Cloud (Cloud Run), and DigitalOcean.
• Native Build Pipeline: Refactored GitHub Actions to use native runners (ubuntu-arm64, macos-arm64), resulting in faster builds and authentic Apple Silicon signing.

Backend & Performance

• Bounded Worker Pool: Implemented a more efficient worker pool with per-user drain logic to prevent message deduplication issues and optimize resource usage.
• Tool-Call Persistence: Enhanced database handling to ensure tool-call metadata is correctly persisted and restored across system reloads.
• Reliable Scheduling: Fixed issues with cron job dispatches and task notification targeting.

Frontend & DX

• FirstBoot Wizard Overhaul: Redesigned the onboarding flow to be fully compatible with the dynamic native plugin system.
• Vite 8 & Rolldown: Upgraded the frontend build system to Vite v8, improving compilation times and reducing the final bundle size.
• Memory UI Refinement: Improved the visualization and deletion of memory relations in the knowledge graph.

Bug Fixes

• Security: Fixed potential prototype pollution vulnerabilities identified via CodeQL.
• Data Integrity: Fixed cross-message leakage between different users in multi-user environments.
• Persistence: Fixed agent.provider and credential overwrite issues in the Settings UI.
• Wizard: Resolved several crashes during the initial configuration of the Docker Model Runner.
• Frontend: Fixed No QueryClient set errors and deduplicated solid-query in production builds.

Refactoring & Code Quality

• Linter Compliance: Massive cleanup of linter warnings and test failures across the Go and TypeScript codebases.
• SDK Consolidation: Migrated plugin implementations to official SDKs (e.g., anthropic-rs, ollama-rs).
• CI/CD Security: Added missing permissions and security hardening to GitHub workflows.

v0.3.0

What's New

Memory System Overhaul

• Map-reducer based memory consolidation: Introduced a new pipeline architecture for consolidating memories, improving scalability and reliability.
• Typed memories: Extended the add_memory tool to support typed memory entries, enabling richer semantic categorisation.
• Extended entity types: Memory consolidation now recognises a broader range of entity types for more accurate knowledge graph construction.
• Improved pipeline reliability: Fixed deduplication logic and enhanced overall robustness of the memory consolidation pipeline.
• OpenClaw backwards compatibility: OpenLobster now supports MEMORY.md to facilitate the migration.

AI Reasoning Level Configuration

• Unified AI reasoning level configuration across all supported providers, making it easier to control model behaviour from a single place.

CLI Improvements

• Added serve, config, migrate, and version subcommands to the binary for a cleaner command-line interface.
• New flags for the serve command: --host, --port, --data-dir, and base_dir support.
• Added migration steps wired into the binary so schema upgrades can be run explicitly.

Permissions & User Management

• Chip-based user selector in the WebUI with bulk-by-group assignment.
• Auto-select agent when creating permissions.
• Fixed provider credential overwrite when editing permission entries.

Pairing Requests

• Added support for pending pairing requests, allowing agents to place incoming pairings on hold until approved.
• Fixed propagation of channelID, channelType, and platformUserName fields during pairing flows.

Task Scheduling

• Improved task_add tool: appended notify username string for cleaner notification targeting.
• Fixed hot-scheduling of new tasks that were not being picked up correctly.

Frontend Improvements

• Replaced hardcoded internal tool list with live API data and polished the general UI.
• Fixed pairing request UI and added hover effects to schema fields.

Bug Fixes

• Fixed group deletion in the WebUI.
• Fixed GetContextWindow returning incorrect data for some service configurations.
• Fixed cross-message leakage between different users.
• Fixed Neo4j node type changes via the WebUI.
• Fixed agent.provider not being persisted and respected on reload.
• Fixed WhatsApp/Twilio snapshot inconsistencies.

Refactoring & Code Quality

• Refactored logging system for consistency across backend services.
• Split main.go wiring into a dedicated cmd/openlobster/init/ package.
• Moved business logic out of main.go into proper internal packages.
• Improved test coverage for both frontend and backend.
• Fixed linter warnings and test failures across the codebase.

v0.2.0

This release consolidates the evolution from the initial structure (v0.1.0) with improvements in stability, implementation refinements, dependency updates, and expanded documentation.

Main changes

• Backend

  • Improvements and fixes across the MCP integration and its GraphQL layer (resolvers, tool handling, and OAuth flow).
  • Refinements to the message handling logic and related processing (including compaction and message processing flow).
  • Infrastructure adapter updates (e.g., terminal handling, memory/Neo4j/GML, filesystem) plus configuration/persistence/migration adjustments.

• Frontend

  • UI/UX adjustments across key views (chat, MCPs, settings, skills, tasks, and memory).
  • Added/updated components to improve rendering and user experience (e.g., skeleton placeholders during loading).
  • Visual/behavioral updates in modals and navigation/header areas.

• Documentation & tooling

  • Review and expansion of architecture and workflow documentation (chat, memory, settings, skills, tasks, etc.).
  • Dependency updates and small infrastructure consistency fixes.

Commits included (since v0.1.0)

• f014cec fix: remove duplicate nodes and fix visibility
• 49a90f5 fix: address review feedback
• 6a3534e docs: improve documentation clarity
• 88a1fd8 deps: update dependencies
• a4876c7 fix: clean up leftover code

Full Changelog: https://github.com/Neirth/OpenLobster/commits/v0.2.0-rebase

Release v0.1.0

This is the initial release of this project. Enjoy!