v1.8.0

Release v1.8.0

v1.8.0-beta.2

What's new in v1.8.0-beta.2

• Agent Todo System: nuevo sistema de tareas para agentes con auto-purga al completarse, desacoplado del ciclo de vida del chat (estado derivado del transcript), spinner sustituido por el logo Levante para tareas en progreso, y herramientas internas (todo_*, present_files) ocultas del UI.
• MCP Apps (SEP-1865): soporte para la extensión MCP UI con widgets renderizados en panel lateral en lugar de inline, y formato anidado de resourceUri.
• Platform resiliente: catálogo de modelos con retry, máquina de estados y feedback de error en UI; auto-selección de modelos flagship curados al sincronizar un proveedor por primera vez.
• Compactación robusta: fallback escalonado para evitar bloqueos cuando el contexto supera el límite del modelo.
• Seguridad: 25 vulnerabilidades resueltas vía pnpm overrides (44 → 19).

✨ Destacados

Este beta consolida la primera versión completa del sistema de todos para agentes, ahora desacoplado del chat y con UI más limpia (sin tool-calls internas visibles). El soporte de MCP Apps desplaza los widgets al panel lateral, mejorando la legibilidad de conversaciones complejas. La capa de plataforma gana resiliencia con una máquina de estados explícita para la carga de modelos y selección automática de los flagships curados, reduciendo fricción al configurar un proveedor nuevo. Cierra el círculo un fallback de compactación que evita el deadlock cuando una conversación crece más allá del límite del modelo.

Features

• feat(models): auto-select curated flagship models on first provider sync
• feat: add agent todo system with auto-purge of completed tasks
• feat: hide internal tool calls (todo_*, present_files) from chat UI
• feat: replace spinner with Levante breathing logo on in-progress todos
• feat(mcp): render MCP App widgets in side panel instead of inline
• feat(mcp): declare MCP Apps (SEP-1865) UI extension and support nested resourceUri format
• feat(platform): add resilient model catalog loading with retry, state machine, and UI error feedback

Bug Fixes

• fix(compaction): add staged fallback to avoid deadlock when context exceeds model limit
• fix(security): resolve 25 vulnerabilities via pnpm overrides (44→19)
• fix(deps): cap @xmldom/xmldom override to ^0.8.12 (unblocks macOS packaging)
• fix: error allowed models

Refactoring & Chores

• refactor(todos): decouple todo system from chat lifecycle via transcript-derived state
• perf: remove devtools middleware from LogViewerStore to fix severe UI degradation
• test: add unit tests for platformService error handling and platformStore state machine
• docs: PDR de conversaciones en background
• chore(landing): mejoras de cards y secciones

v1.7.5

What's new in v1.7.5

• Fixed model resolution for allowed models in Levante Platform

✨ Destacados

Simplified the model target resolver by removing client-side allowed model validation. The platform server now handles model plan filtering directly, eliminating redundant checks and fixing errors when selecting models in platform mode.

Bug Fixes

• fix: resolve error with allowed models validation in platform mode — removed client-side isPlatformModel checks, delegating model availability validation to the server

Refactoring & Chores

• refactor: simplified modelTargetResolver.ts by removing ~40 lines of redundant platform model validation logic
• Updated tests to reflect the simplified resolution flow

v1.7.4

What's new in v1.7.4

• Skills system with built-in defaults, custom .zip imports, and .skill packages
• PDF viewer with continuous scroll replacing page-by-page navigation
• File watcher for auto-updating sidebar file tree
• HTML panel rendering and Chat/Cowork mode selector
• Per-message model usage analytics tracking

✨ Destacados

This release brings the new Skills system — Levante now ships with built-in default skills (including a PDF viewer and a skill creator), supports importing custom skills from .zip files and .skill packages, and introduces the present_files tool. The PDF viewer has been redesigned with continuous scrolling for a smoother reading experience. A file watcher now keeps the sidebar tree in sync automatically, and self-contained HTML files can be rendered directly in the side panel.

Features

• feat(skills): add default skill-creator seeding on app startup
• feat(skills): add custom skill import from .zip file
• feat(skills): add pdf as default bundled skill
• feat(skills): prevent removal of built-in default skills in UI
• feat(skills): add present_files tool and .skill package support
• feat(pdf): replace page-by-page navigation with continuous scroll viewer
• feat(fs): add file watcher for auto-updating sidebar file tree
• feat(panel): render self-contained HTML files in side panel via srcDoc iframe
• feat(analytics): replace provider_stats with per-message model_usage tracking
• feat(projects): add files button in sidebar and fix i18n in ProjectPage
• feat(layout): show project name below chat title when sidebar is open
• feat(ProjectPage): add Chat/Cowork mode selector
• feat: allow platform users to use external providers alongside platform models

Bug Fixes

• fix(ui): prevent overflow in background tasks dropdown layout
• fix(skills): rebrand eval viewer messages from Claude Code to Levante
• fix(skills): update default skill-creator with Levante/Cowork corrections

v1.7.4-beta.2

What's new in v1.7.4-beta.2

• Fixed allowed models validation error

✨ Destacados

Corrección de un error en la validación de modelos permitidos que podía impedir el uso correcto de ciertos modelos en la aplicación.

Bug Fixes

• fix: error allowed models

v1.7.4-beta.1

What's new in v1.7.4-beta.1

• Skills ecosystem: Custom skill import, .skill packages, built-in PDF skill, and skill-creator seeding
• File system: Real-time file watcher auto-updates the sidebar file tree
• Panel viewer: Self-contained HTML files now render in the side panel via srcDoc iframe
• PDF viewer: Continuous scroll replaces page-by-page navigation
• Analytics: Per-message model usage tracking replaces aggregate provider stats
• Projects: Files button in sidebar, Chat/Cowork mode selector, and project name in chat header

✨ Destacados

This release significantly expands the skills platform with custom skill import from .zip files, a built-in PDF skill, and automatic skill-creator seeding on startup. The file system now features a real-time watcher that keeps the sidebar tree in sync. The PDF viewer has been redesigned with continuous scrolling for a smoother reading experience, and analytics now track model usage per message for finer-grained insights.

Features

• feat(fs): add file watcher for auto-updating sidebar file tree (#235)
• feat(panel): render self-contained HTML files in side panel via srcDoc iframe
• feat(skills): add present_files tool and .skill package support
• feat(pdf): replace page-by-page navigation with continuous scroll viewer
• feat(analytics): replace provider_stats with per-message model_usage tracking
• feat(skills): add pdf as default bundled skill
• feat(skills): prevent removal of built-in default skills in UI
• feat(skills): add default skill-creator seeding on app startup
• feat(skills): add custom skill import from .zip file
• feat(projects): add files button in sidebar and fix i18n in ProjectPage
• feat(layout): show project name below chat title when sidebar is open
• feat(ProjectPage): add Chat/Cowork mode selector
• feat: allow platform users to use external providers alongside platform models

Bug Fixes

• fix(ui): prevent overflow in background tasks dropdown layout
• fix(skills): rebrand eval viewer messages from Claude Code to Levante
• fix(skills): update default skill-creator with Levante/Cowork corrections

v1.7.3

What's new in v1.7.3

OAuth subscriptions ahora soportan OpenAI Codex, se resuelven los fallos de auto-envío con tool calls paralelas, y se refinan las skills en modo Cowork.

✨ Destacados

Esta versión introduce soporte completo para OAuth de suscripción con OpenAI Codex, convirtiendo el flujo anterior exclusivo de Anthropic en un sistema genérico y extensible. Se corrige un bug crítico donde el chat enviaba peticiones antes de que todos los tool calls paralelos hubieran sido aprobados. Además se mejora la gestión de skills vinculándola al modo Cowork efectivo y se depuran duplicados en el selector.

Features

• feat(oauth): Sistema OAuth genérico para suscripciones con soporte para OpenAI Codex — reemplaza el handler específico de Anthropic con un SubscriptionOAuthService reutilizable y configurable por proveedor

Bug Fixes

• fix(chat): Evita el auto-envío cuando hay aprobaciones de tool calls paralelas pendientes — añade toolMessageSanitizer y utilidad toolApprovalAutoSend
• fix(skills): Vincula disponibilidad de skills al modo Cowork efectivo
• fix(skills): Deduplica skills en el selector cuando están instaladas en ambos scopes
• fix(auto-update): Corrige la entrega de actualizaciones en macOS ARM64
• fix(ci): Fuerza shell bash en el paso de upload de artefactos en Windows

Refactoring & Chores

• style(theme): Paleta de modo claro actualizada a tonos cálidos (sand/pearl)
• style(sidebar): Reorganización del layout de la sección de proyectos en el sidebar
• Mejoras en la página de modelos al hacer login
• Actualización de dependencias

v1.7.3-beta.1

What's new in v1.7.3-beta.1

OAuth subscriptions ahora soportan OpenAI Codex, se resuelven los fallos de auto-envío con tool calls paralelas, y se refinan las skills en modo Cowork.

✨ Destacados

Esta versión introduce soporte completo para OAuth de suscripción con OpenAI Codex, convirtiendo el flujo anterior exclusivo de Anthropic en un sistema genérico y extensible. Se corrige un bug crítico donde el chat enviaba peticiones antes de que todos los tool calls paralelos hubieran sido aprobados. Además se mejora la gestión de skills vinculándola al modo Cowork efectivo y se depuran duplicados en el selector.

Features

• feat(oauth): Sistema OAuth genérico para suscripciones con soporte para OpenAI Codex — reemplaza el handler específico de Anthropic con un SubscriptionOAuthService reutilizable y configurable por proveedor

Bug Fixes

• fix(chat): Evita el auto-envío cuando hay aprobaciones de tool calls paralelas pendientes — añade toolMessageSanitizer y utilidad toolApprovalAutoSend
• fix(skills): Vincula la disponibilidad de skills al modo Cowork efectivo
• fix(skills): Deduplica skills en el selector cuando están instaladas en ambos scopes
• fix(auto-update): Corrige la entrega de actualizaciones en macOS ARM64
• fix(ci): Fuerza shell bash en el paso de upload de artefactos en Windows

Refactoring & Chores

• style(theme): Paleta de modo claro actualizada a tonos cálidos (sand/pearl)
• style(sidebar): Reorganización del layout de la sección de proyectos en el sidebar
• Mejoras en la página de modelos al hacer login
• Actualización de dependencias

v1.7.2

What's new in v1.7.2

• Levante Platform with OAuth: Sign in to Levante Platform directly from the app, with Claude Max/Pro subscription support and authenticated model access
• Projects: Organize chat sessions into projects with dedicated pages, model selectors, and scoped sidebar history
• Cowork Mode & Coding Tools: New Cowork mode with working directory selection, tool approval flow, diff previews, and background task execution
• MCP Code Mode: Enable AI agents to write JavaScript that orchestrates multiple MCP tools, with fullscreen widget fixes
• Skills System: Install and toggle skills per chat from a dedicated panel, with project-scoped skill downloads and a Skills Store
• Cross-platform improvements: NSIS installer for Windows, Linux AppImage support, and macOS keychain fixes

✨ Destacados

Levante v1.7.2 introduces Levante Platform as a first-class app mode with full OAuth 2.1 authentication, letting users sign in and access Claude Max/Pro subscriptions directly from the sidebar. The new Projects system brings organization to your workflow by grouping chat sessions with dedicated project pages and per-project model selection. Cowork Mode adds a complete coding assistant experience with tool approval, inline diff previews, background tasks, and a web preview panel with automatic port detection. Under the hood, the logging system has been migrated to Winston, the installer stack now covers Windows (NSIS), macOS, and Linux, and dozens of stability fixes improve the day-to-day experience.

Features

Platform & Authentication

• feat(platform): implement platform mode as first-class app mode with OAuth
• feat(anthropic): implement Claude Max/Pro OAuth subscription and fix reasoning panel
• feat(platform): show authenticated state on Levante Platform card
• feat(platform): obtain orgId from /api/v1/me when not present in the JWT
• feat(sidebar): move platform sign-in button to sidebar and simplify feedback
• feat: add Levante Platform as OAuth-based model provider

Projects & Organization

• feat: add projects feature for organizing chat sessions
• feat(projects): auto-create project directory and fix dropdown event propagation
• feat(projects): add model selector on project page
• feat(ui): show project-scoped chat history in sidebar
• feat(ui): add delete chat in project page

Chat & UI

• feat(chat): add file mentions with @ autocomplete in chat input
• feat(autocompact): add context budget tracking, token usage, and conversation compaction
• feat(ui): redesign sidebar with collapsible projects and inline actions
• feat(ui): add PDF preview with streaming protocol and pdfjs-dist
• feat(ui): add file browser panel and unified side panel
• feat(chat): classify streaming errors and show friendly messages
• feat(models): show "Private" badge for zero_data_retention models
• feat(file-browser): add backend-powered file search in sidebar Files tab

Cowork Mode & Tools

• feat: add coding tools integration with Cowork mode
• feat(cowork): add working directory selector for Cowork mode
• feat: add background tasks system for long-running commands
• feat(approval): restore full tool approval flow with approve/deny UI
• feat(approval): add configurable tool approval per provider with UI warning
• feat(diff): show filename in inline diff and add diff preview to tool approval
• feat: web preview side panel with automatic port detection

MCP & Code Mode

• feat(mcp): implement Code Mode and fix MCPSession API
• feat(mcp): add individual tool enable/disable functionality

Skills & Store

• feat(skills): add Skills panel to chat UI with per-skill toggle
• feat: inject installed skills into agent context via skill_execute tool
• feat(store): separate installed and store views for MCPs and Skills

Infrastructure

• feat(windows): migrate installer from Squirrel to NSIS
• feat(release): add Linux AppImage support
• feat: migrate logging system from custom transports to Winston
• refactor(env): centralize environment variables in EnvConfig

Bug Fixes

• fix(oauth): store tokens in plaintext instead of keychain to avoid repeated prompts
• fix(platform): clean OAuth state on logout
• fix(mcp): fix fullscreen for MCP Apps widgets (SEP-1865)
• fix(chat): enable streamdown code block copy and download buttons
• fix(chat): use active chat model for title generation
• fix(ai): prevent asterisks in auto-generated chat titles
• fix(approval): handle tool denial without infinite loop
• fix(database): repair projects schema and harden migration runner
• fix(models): resolve race conditions in model service initialization and sync
• fix(config): default to production URLs when no env vars are set
• fix(macos): add keychain entitlements to prevent repeated password prompts
• fix(windows): handle Squirrel events to create shortcuts correctly
• fix(build): bundle jwt-decode and diff in main process
• fix(release): fix missing assets in GitHub releases

Refactoring & Chores

• perf(oauth): reduce post-login latency
• refactor(approval): simplify tool approval UI and add i18n support
• chore(deps): phased dependency update (security + minor/patch + major)
• chore(deps): upgrade mcp-use from 1.11.2 to 1.13.4
• security: fix 9 Dependabot alerts

v1.7.2-beta.5

v1.7.2-beta.5