Add ToolSelector for enabling sandboxed python per agent, drop unsafe
filesystem/bash tools, bundle fetchUrl with web search, and return
Monty's last expression value so agents need not print.
remarkSplitBlocks was splicing after-break paragraphs onto the root
tree, which dropped content inside list items and other nested parents.
Insert the split sibling into the actual parent instead.
Also store timestamps as timestamptz so Drizzle no longer treats naive
timestamp values as UTC and shifts displayed times by the server offset.
- Add HMAC-based file token auth for secure AI model file access
- Add file upload retry with exponential backoff (max 3 retries)
- File endpoint now requires session auth or signed token
- Support assistant role messages in chat input
- Optimistic UI for attachments on message send
- Verify topic ownership before allowing messages
- Switch web scraping to Firecrawl API
- Agent profile page layout fixes (proper flex overflow)
- Add quick switcher (Ctrl+K) to sidenav
- Clean up longcat.ts and stale comments
- Replace Triplit env vars with PostgreSQL, auth, and S3 config
- Add app service to docker-compose with healthcheck dependency
- Bind ports to localhost only for security
- Add required POSTGRES_PASSWORD validation
- Add scripts/start.ts to run drizzle migrations before server start
- Update Dockerfile to copy migration files and install prod deps
- Change entrypoint to use migration-aware startup script
- Add deployment documentation
- Add HuggingFace as AI provider with model fetching
- Add Poolside, Tencent, and HuggingFace logo components
- Update model mappings for new providers
- Add @ai-sdk/huggingface dependency
- Bump longcat-ai-sdk-provider to 0.0.5
- Add /resources page with file listing, filtering, and bulk operations
- Add useResources composable for file state management
- Add sidenav header and nav components for resources section
- Add files list API endpoint (GET /api/files)
- Track file size during upload
- Add resources link to home navigation
- Replace focusedIndex integer with activeChildId string in messages schema
- Add migration to convert existing data and drop focused_index column
- Add hooks for providers.updatedAt and messageParts.lastUpdatedAt
- Add file size column to files table
- Update message navigation to use child ID-based lookup
- Simplify delete logic by removing focusedIndex math
- Fix file delete endpoint filename ([id[ -> [id])
Fixes link color from hardcoded white to currentColor for proper theme
support. Adds video and audio modality icons to ModelInfo. Registers
anthropic, nvidia, and xiaomi in the provider icon map. Removes unused
pagination validation from agents and topics endpoints. Adds convenience
start script. Updates AGENTS.md.
Awaits sendMessage before navigating to prevent race conditions. Passes
agentId to startGeneration for correct agent resolution. Adds topic
deletion via Ctrl+Alt+Backspace shortcut. Adds date/time prefix to user
messages in LLM context. Fixes reasoning-container typo.
Adds remark-breaks for newline-to-br conversion and a custom
remarkSplitBlocks plugin to split paragraphs at hard breaks. Code blocks
over 15KB are truncated with a 'Show all' toggle for better
performance.
Sidenav becomes a full-screen overlay on mobile with slide-in animation.
Dialog goes full-screen on small screens. Settings dialog gets a mobile
nav with hamburger menu. Touch targets and text sizes scaled up for
mobile. Slider gets visual checked/unchecked colors. Sidebar auto-closes
on mobile route navigation. Resize handle hidden on mobile.
Full-featured image viewer with pinch-zoom, scroll-zoom, drag-pan,
double-click reset, and keyboard shortcuts. Animated open/close
transitions match the origin element position. Integrates into
attachment display with click-to-open behavior.
Integrates SearXNG web search as a tool available during chat when the
agent has search enabled. Supports optional reranking of results via a
configurable reranking model. Replaces Python subprocess evaluation with
@pydantic/monty WASM runtime. Introduces stable tool call ID mapping to
avoid exposing provider-native IDs to the database.
Adds a config jsonb column to agents for per-agent search settings
(enabled, rerank, maxResults). Types systemAssistants with rename and
rerank sub-objects. Backfills missing systemAssistants keys on settings
fetch. Includes generated drizzle migrations.
Each AI provider now has its own module implementing a ProviderModule
interface with createGateway, getAuthHeaders, and fetchModels methods.
This replaces the monolithic switch statements in ai-provider.ts and
models.post.ts with a clean registry pattern.
zoeissleeping
