KIRA 5.0 offline local AI agent

told yall I have a lot in mind. This is just the start. Get KIRA now while its only $25 to own for life. all future updates included.

Here's a complete inventory of everything built in Kira 5.0: 3 Main Modes

1. Chat Mode Traditional conversational AI interface Message history, image preview (drag-and-drop), webcam capture, screen capture Microphone input (voice-to-text), text-to-speech output New Chat, Learn, Wipe Memory buttons
2. Agent Mode Autonomous assistant with screen monitoring ("KIRA is watching") Pulsing eye animation UI Floating chat widget (top-right corner) with minimize Proactive help based on observing your work
3. Code Mode (Full IDE) Monaco Editor — syntax highlighting, tabs, minimap, auto-completion, code folding File Explorer — recursive directory tree, open folder dialog, click-to-open Code Search — project-wide search with clickable results (line numbers) Git Panel — status, diff, commit, branch display Terminal — xterm.js + node-pty, multiple terminal tabs, resize, PowerShell/bash Preview Panel — live HTML/web preview iframe, refresh, open in browser Welcome Screen — Start New Project, Open Folder, Open File, keyboard shortcuts reference Activity Bar (Left Sidebar in Code Mode)

Explorer (Ctrl+B), Search (Ctrl+Shift+F), Source Control (Ctrl+Shift+G), KIRA Chat (Ctrl+Shift+K) 3 Auxiliary Agent buttons: Claude (CLD), ChatGPT (GPT), Gemini (GEM) Mode switcher button KIRA Chat Panel (in Code Mode)

Backend connection status indicator + Restart Backend button Toolbar buttons: New Chat, Help, Status, Models, Usage, Test, Git, Search, Web, Auto (Autonomous Mode), War Room, Clear, Learn, Wipe Streaming responses, code blocks with syntax highlighting, markdown rendering Image drag-and-drop upload, webcam (snap + tracking mode at 3s/5s/10s/15s intervals), screen capture Voice input + TTS buttons Stop button during streaming Auxiliary Agent Panels (Claude, ChatGPT, Gemini)

Independent message histories per agent Real-time streaming display Availability indicators + model display Separate input areas per agent All aux agents get full tool access (same tools as KIRA) Multi-Provider LLM Backend

ProviderDefault ModelAlso AvailableKIRA Brain (Ollama, offline)llama3:latestllama3.2-vision, nomic-embed-textClaudeclaude-haiku-4-5claude-sonnet-4-5, claude-opus-4-6ChatGPTgpt-4o-minigpt-4.1-mini, gpt-4o, gpt-4.1, o3-miniGeminigemini-2.0-flashgemini-2.5-flash, gemini-2.5-pro Runtime model switching, per-model pricing database, tier classification (cheap/mid/premium) Daily budget caps per provider ($2 Claude, $1 OpenAI, unlimited Gemini) Budget warnings at 75% threshold, token usage tracking Tools Available to All Agents

File ops: read_file, write_file, edit_lines, create_directory, list_files Terminal: run_command (async, 120s timeout, blocked command filtering) Search: search_files (ripgrep-based, regex, ignores node_modules/.git) Git: git_status, git_diff, git_commit, git_log Web: web_search (DuckDuckGo), web_fetch, web_browse (search + read top 3) Testing: run_tests (auto-detects pytest/jest/mocha/cargo/go) Media: generate_image (Stable Diffusion + DALL-E fallback), generate_video (ffmpeg), generate_diagram (Mermaid) Advanced Modes

Autonomous Mode Goal-driven self-directed work (up to 25 steps) Self-evaluates progress with STATUS markers (CONTINUE, DONE, BLOCKED) Step-by-step progress UI War Room Sends problem to ALL agents simultaneously (KIRA + Claude + ChatGPT + Gemini) Concurrent execution, KIRA synthesizes the best answer from all responses Error Watcher Toggle button (EW) on terminal header Auto-detects terminal errors via regex patterns Captures error context, sends to KIRA for auto-diagnosis and fix Slash Commands

/mode, /status, /clear, /test, /search, /grep, /git, /web, /browse, /learn, /model, /models, /usage, /help Persistent Memory & Database (SQLite — kiradex.db)

All data persists on disk across sessions indefinitely. Nothing is wiped automatically. conversations — all user/assistant messages stored permanently, session_id for organizational grouping, timestamped knowledge_blocks — deduplicated by content hash, importance scoring, access counting, categorical tagging memory_records — reinforcement scoring, success/failure tracking, avoid markers, categorical tagging vec_knowledge / vec_records — 768-dimensional semantic embeddings (nomic-embed-text) for similarity search knowledge_fts / records_fts — full-text search indexes agent_state — persistent key-value store for agent state schema_version — migration tracking Wipe Memory is the only way to clear data (12h, 24h, 48h, or all) Other Features

License system — activation, validation, offline grace period Webcam tracking — continuous capture at configurable intervals Screen analysis — vision model analyzes screenshots with custom prompts Keyboard shortcuts — Ctrl+B, Ctrl+`, Ctrl+Shift+K, Ctrl+S, Ctrl+P, Ctrl+N, Ctrl+Shift+P Resizable sidebar and terminal (drag handles) Dark theme UI (#1e1e1e) WebSocket server on port 7855, multi-client support, 50MB message limit for images Python backend auto-launch — Electron spawns and manages the Python process, port conflict resolution, graceful shutdown