axpdev-lab/aeroftp v2.0.0

AeroFTP v2.0.0

on GitHub

[2.0.0] - 2026-02-07

AeroAgent Pro — Professional AI Experience

AeroAgent evolves from a capable assistant into a professional-grade AI development companion with 5 enhancement phases: Provider Intelligence, Advanced Tool Execution, Context Intelligence, Professional UX, and Provider-Specific Features.

Added

• Streaming markdown renderer: Incremental rendering with finalized segments (React.memo, never re-rendered) and in-progress tail. Eliminates flashing during AI responses
• Code block actions: "Copy", "Apply", "Diff", and "Run" buttons on every code block in AI responses. "Apply" writes code to the active editor file, "Diff" shows side-by-side comparison before applying
• Agent thought visualization: ThinkingBlock component shows Claude/OpenAI/Gemini reasoning process with duration timer, token count badge, and collapsible content
• Prompt template library: 15 built-in templates activated with / prefix — /review, /refactor, /explain, /debug, /tests, /docs, /security, /optimize, /fix, /convert, /commit, /summarize, /types, /analyze-ui, /performance. Custom templates storable in vault
• Multi-file diff preview: PR-style diff panel for reviewing changes across multiple files with per-file checkboxes and unified apply action
• Cost budget tracking: Per-provider monthly spending limits with warning thresholds, conversation cost display, and vault-persisted spending records
• Chat search: Ctrl+F search overlay with role filtering (all/user/assistant), match highlighting, and keyboard navigation between results
• Keyboard shortcuts: Ctrl+L (clear), Ctrl+Shift+N (new chat), Ctrl+Shift+E (export), Ctrl+F (search), Ctrl+/ (focus input)
• Anthropic prompt caching: System messages sent with cache_control: {"type": "ephemeral"} — cache reads are 90% cheaper, with savings displayed per message in cyan
• OpenAI structured outputs: strict: true on function definitions with additionalProperties: false for OpenAI, xAI, and OpenRouter providers — ensures reliable tool call JSON schemas
• Ollama model-specific templates: 8 model family profiles (llama3, codellama, deepseek-coder, qwen, mistral, phi, gemma, starcoder) with tailored prompt styles and optimal temperatures
• Ollama pull model from UI: Text input + progress bar in AI Settings to download models directly via POST /api/pull with real-time NDJSON streaming progress
• Ollama GPU memory monitoring: GPU Monitor panel in AI Settings shows running models, VRAM usage with color-coded bars, and auto-refresh every 15 seconds
• Gemini code execution: Parse and render executableCode and codeExecutionResult response parts with syntax-highlighted code blocks and collapsible output sections
• Gemini system_instruction: System prompt now passed as top-level systemInstruction field instead of in-message, following Google API best practices
• Gemini context caching: New gemini_create_cache command for caching large contexts (32K+ tokens) with configurable TTL — reduces latency and cost by up to 75% on subsequent requests
• Thinking budget presets: 5 presets (Off/Light/Balanced/Deep/Maximum) plus range slider (0-100K tokens) for fine-grained control of AI reasoning depth
• Provider Intelligence Layer: Per-provider system prompt profiles with optimized identity, style, and behavior rules for all 7 AI providers
• DAG-based tool pipeline: Topological sort of tool calls by path dependencies — independent tools execute in parallel, dependent tools run sequentially
• Diff preview for edits: Read-only diff preview in tool approval UI for local_edit and remote_edit tools (100KB cap)
• Intelligent tool retry: analyzeToolError() with 8+ error detection strategies and automatic retry suggestions
• Tool argument validation: Pre-execution validation via validate_tool_args Rust command — checks file existence, permissions, dangerous paths, and size limits
• Composite tool macros: Reusable multi-tool workflows with {{var}} template variables, max depth 5, new "Macros" tab in AI Settings
• Tool progress indicators: Real-time progress bars for long-running tools (upload, download, RAG indexing) via ai-tool-progress Tauri events
• Project-aware context: Auto-detect project type from 10 markers (Node.js, Rust, Python, PHP, Go, Java, Maven, Gradle, CMake, Make) — injects metadata into AI system prompt
• File dependency graph: scan_file_imports parses import/require/use statements in 6 languages (JS/TS, Rust, Python, PHP, Go, Java) for context-aware suggestions
• Persistent agent memory: .aeroagent file per project — AI reads at session start, can write learnings via agent_memory_write tool for cross-session knowledge retention
• Conversation branching: Fork conversations at any message to explore alternative approaches. Branch selector dropdown with create, switch, and delete operations
• Smart context injection: analyzePromptIntent() detects task type from user prompt and auto-selects relevant context (git diff, file imports, project info, agent memory) with priority allocation
• Token budget optimizer: Dynamic allocation based on model capacity — full/compact/minimal modes with visual segmented bar showing system/context/history/current/available token breakdown
• Universal Preview syntax highlighting: Prism.js-powered source code coloring for 25+ file types in the preview modal, with language badge and full text selection support
• HTML/Markdown render toggle: Preview modal now offers live rendering for HTML files (iframe with inlined local CSS) and Markdown files (via MarkdownRenderer), plus responsive viewport controls (mobile/tablet/desktop), zoom slider, and browser-open action
• Image color picker: Canvas-based pixel color sampling in ImageViewer (cross-platform), with hex value display and clipboard copy

Changed

• AeroAgent tool count: Expanded from 27 to 28 tools with addition of agent_memory_write
• AI Settings tabs: Reorganized from 6 to 7 tabs with new "Macros" tab
• Anthropic API version: Unified to 2025-04-15 for all Anthropic calls (caching + thinking support)
• System prompt architecture: Provider-aware profiles replace one-size-fits-all prompt. Ollama models get family-specific guidance
• Token info display: Message footer now shows cache savings (cyan arrow-down icon) when Anthropic prompt caching reduces costs
• Settings reorganization: Lock Screen Pattern moved from Security to Appearance tab; Vault Backup moved to dedicated Backup tab with Key icon

Fixed

• Chat search navigation: data-message-id attributes enable scroll-to-match functionality
• Cost tracking accuracy: Budget check runs before sending messages, preventing overspend
• Template detection: / prefix in chat input triggers template selector popup without false positives on regular text
• Streaming token capture: Cache creation and cache read tokens now correctly captured from streaming chunks
• Gemini system prompt: Previously sent as first user message, now correctly uses systemInstruction top-level field
• Auto-lock timeout persistence: Timeout was only stored in RAM — now persisted to config file and restored on app restart
• Auto-lock slider default: Fixed slider showing 5 minutes when disabled (now correctly shows 0)
• Auto-lock save flow: Removed broken "Save Timeout" button, timeout now auto-saves on slider change
• EyeDropper color picker on Linux: Hidden when EyeDropper API is unavailable (WebKitGTK), preventing runtime errors

Downloads:

• Windows: .msi installer or .exe
• macOS: .dmg disk image
• Linux: .deb, .rpm, .snap, or .AppImage