application-feature-list.md

Chitty Workspace Feature List and Usage Guide

This document inventories the feature sets that currently exist across the Chitty Workspace application, including the desktop app, embedded web UI, local HTTP server, browser extension, CLI entry points, and bundled local model sidecar support. It is organized by feature set so a user can quickly understand what each section of the product is for and how each feature is used.

Feature Sets Overview

Feature Set	Overall Description
Application Shell	The desktop shell starts the local server, opens the WebView UI, and keeps the app available from the system tray.
Chat Workspace	The main multi-panel workspace where users open chat panels, pick agents/models, set project context, and talk to the assistant.
Agents and Skills	User-defined or AI-generated agents combine persona, skills, model preferences, approval settings, and project scope.
Tools and Automation	Native, custom, integration, marketplace, and browser-driven tools let the assistant search code, read/write files, use the terminal, browse, and load skills.
Providers and Models	Chitty supports cloud BYOK providers and a built-in local engine for GGUF inference, image/video/TTS generation, speech-to-text (Whisper), and LoRA fine-tuning.
Marketplace and Connections	Installable packages expand the system with tools, optional setup steps, and background connections such as event listeners.
Memory and Context	Persistent memory, project context files, token usage, and context-budget controls help the assistant retain useful information while staying within model limits.
Scheduling and Background Services	The app can store recurring scheduled tasks and runs server-side background loops for schedules and package connections.
Browser Extension	The browser extension reports connection status and enables browser automation commands against a real Chrome or Edge session.
CLI and Local Services	Command-line entry points support starting the app, inspecting configuration, listing agents, and running limited tests; the built-in sidecar enables local inference, media generation, and training.

Detailed Feature List

Feature Set	Feature	What it does	How to use
Application Shell	Desktop launch / default run mode	Starts storage, configuration, tool runtime, local server, and the embedded WebView UI.	Run chitty-workspace with no subcommand. The app launches the local HTTP server and opens the desktop window automatically.
Application Shell	System tray presence	Keeps the app available from the tray even when the main window is closed.	Start the app, then use the tray icon menu to reopen the window with Open or fully exit with Quit. Closing the window hides it to the tray instead of exiting.
Application Shell	Local embedded web UI	Loads the app UI from the local server into a native desktop window.	Launch the desktop app normally; the WebView automatically points at the local http://127.0.0.1:8770 server.
Chat Workspace	Multi-panel workspace	Lets users work in multiple side-by-side chat panels and switch focus between panels.	Click + New Panel in the top bar to create another panel. Use panel focus and the carousel navigation to move between open panels.
Chat Workspace	Theme toggle	Supports dark and light themes for the UI.	Use the top-bar theme button to switch between dark and light mode.
Chat Workspace	Per-panel chat	Each panel acts as a separate chat surface with its own messages, input area, agent, model, and project path.	Open a panel, type into the message box, optionally choose an agent/model, and send your prompt.
Chat Workspace	Project path prompt and field	Associates a project directory with a panel so the assistant can use project-scoped context and memories.	Enter a filesystem path in the project prompt or project path field shown in a panel header.
Chat Workspace	Slash commands	Supports command-style workflows such as scheduling and help inside chat.	Type / in chat, then run commands such as /schedule, /schedules, or /help.
Chat Workspace	Action panel	Provides a right-side workspace for secondary views such as agents, providers, marketplace, browser, builder, and activity.	Open the action panel and click the relevant tab to manage agents, providers, marketplace packages, browser tools, or activity details.
Chat Workspace	Activity log and tool call cards	Shows tool execution history, status, duration, and expandable details inline with chat/activity.	Send a task that uses tools, then expand the tool call blocks or review the activity panel.
Chat Workspace	Approval cards	Prompts the user before sensitive actions such as file writes, terminal use, or browser actions when approval is required.	When an approval card appears, choose approve, auto-approve for the session if offered by the workflow, or deny.
Chat Workspace	Artifact cards	Renders generated artifacts in chat so users can inspect outputs the agent produced.	Click the artifact card that appears in a conversation after the assistant emits an artifact.
Agents and Skills	Built-in Chitty agent	Provides the default system administrator-style assistant that understands the platform.	Open a chat panel and use the default assistant without creating a custom agent.
Agents and Skills	Custom agent management	Create, edit, list, and delete agents with their own persona, skills, tags, project scope, provider/model preferences, and execution settings.	Open the Agents area from the action panel or use the API/UI agent forms to create or update an agent, then select that agent in a chat panel.
Agents and Skills	Agent execution configuration	Agents can define max iterations, temperature, max tokens, approval mode, and context controls.	Configure these values when creating or editing an agent, then use that agent for future chats.
Agents and Skills	Agent Builder	Provides an AI-assisted workflow for generating agents conversationally and previewing the resulting draft before saving.	Open the Agent Builder view in the action panel, describe the kind of agent you want, answer follow-up questions, review the preview, and save the draft.
Agents and Skills	Skills catalog and on-demand loading	Discovers SKILL.md-based skills from bundled, user, project, and marketplace locations and loads full instructions only when needed.	Let an agent use discovered skills automatically, or explicitly load a skill through the skill tooling when deeper instructions are needed.
Tools and Automation	Native file tools	Lets the assistant read and write files in the local workspace.	Ask the assistant to inspect or update a file; when approval is required, approve the requested action.
Tools and Automation	Native terminal tool	Runs shell commands from the assistant workflow.	Ask the assistant to execute a command or workflow; review and approve the terminal action if the current agent is in prompt mode.
Tools and Automation	Native code search tool	Searches project code and files for relevant symbols or content.	Ask the assistant to search the codebase, function names, or text across the project.
Tools and Automation	Memory save tool	Stores useful user, project, feedback, or reference memories for later retrieval.	Ask the assistant to remember a preference, project rule, or reusable detail.
Tools and Automation	Tool creation tool	Supports creating additional custom tools.	Use the assistant to generate a new custom tool when a repeated workflow should become reusable.
Tools and Automation	Package install tool	Installs marketplace packages that provide new tools.	Use the marketplace UI or agent workflow to install a package, then complete any required setup steps.
Tools and Automation	Browser automation tool	Sends browser commands through the browser bridge so an agent can navigate, click, type, read, execute JavaScript, and capture state.	Install the browser extension, keep Chitty running, then ask the assistant to perform browser work. Approve browser actions when prompted.
Tools and Automation	Google productivity tools	Includes native Gmail read/send, Calendar list, and Drive search tool registrations.	Connect or configure the required Google authentication path, then ask the assistant to read mail, send mail, inspect calendars, or search Drive.
Providers and Models	BYOK cloud providers	Supports OpenAI, Anthropic, Google AI, and xAI using user-supplied API keys.	Open Settings or the Providers panel, save the provider API key, discover models if needed, then choose the provider/model in chat.
Providers and Models	Local model engine	Runs GGUF models locally via the built-in Python sidecar (llama.cpp). No external services needed.	Place GGUF files in a scan directory, open Settings > Providers > Local Models, scan, and load a model. Select it in the model dropdown to chat.
Providers and Models	Local image generation	Generates images locally using Flux, SDXL, SD3, and more via diffusers.	Register a local diffusion model, load it, then ask the assistant to generate an image. Set image_provider: local in System Defaults.
Providers and Models	Local video generation	Generates videos locally using CogVideoX, Wan, LTX-Video, AnimateDiff, Mochi.	Register a local video model, load it, then ask the assistant to generate a video.
Providers and Models	Local text-to-speech	Converts text to speech locally using Bark, XTTS, SpeechT5, Parler.	Register a local TTS model, load it, then ask the assistant to generate speech.
Providers and Models	Local speech-to-text (Whisper)	Transcribes audio to text using Whisper models. Auto-downloads on first use.	Send audio to the STT endpoint or use the speech-to-text tool. Supports transcription and translation.
Providers and Models	LoRA/QLoRA fine-tuning	Fine-tune HuggingFace models with LoRA/QLoRA directly in the app. Upload datasets, configure training, monitor live progress.	Open Settings > Providers > Fine-Tuning > Open Workbench. Upload a JSONL/CSV dataset, enter a HuggingFace model ID, configure LoRA parameters, and start training.
Providers and Models	Secure key storage	Stores API keys in the OS keyring rather than plain-text config files.	Save a provider key from the UI; Chitty stores it through the platform keyring automatically.
Providers and Models	Model discovery and user model library	Lets users discover models from providers, add them to their local list, remove them, and set defaults.	Use the provider controls to discover models, then add desired models and optionally mark one as the default for that provider.
Marketplace and Connections	Bundled marketplace packages	Seeds bundled marketplace packages such as web tools, social media, Slack, Google Gmail, Google Calendar, and Google Cloud into the local marketplace directory.	Open Marketplace, inspect the available packages, install or enable the package you need, then use its tools from chat.
Marketplace and Connections	Package metadata and setup steps	Packages can expose descriptions, tools, vendor metadata, and installation/setup guidance.	From the marketplace, open a package card, review its tools and setup requirements, and complete any required install steps.
Marketplace and Connections	Background package connections	Long-running package connections can run as subprocesses, send events, heartbeat, and restart on failure.	Install a package that declares connections, enable required features/credentials, and let the background connection manager keep it running.
Memory and Context	Persistent memory types	Stores user, feedback, project, and reference memories in SQLite for reuse across sessions.	Ask the assistant to remember something important, or rely on the app to persist relevant memory created during use.
Memory and Context	Project context files	Loads chitty.md or .chitty/chitty.md files from a project so the assistant understands local conventions and workflows.	Add a chitty.md file to your project, then set that project path in a panel before chatting.
Memory and Context	Context budget controls	Agents can define context budget percentage, compaction strategy, and maximum conversation turns to control long-running conversations.	Set these options on an agent, then use that agent for chats that may become long or tool-heavy.
Memory and Context	Token usage analytics	Tracks token usage by conversation, provider, and model and exposes usage views in the Settings modal.	Open Settings and switch to the Usage tab to review token totals and conversation usage.
Memory and Context	Context usage indicator	Shows a token/context usage badge and progress bar in the chat UI.	Watch the context badge near the model controls and the panel progress indicator during longer conversations.
Scheduling and Background Services	Scheduled task CRUD	Lets users create, list, edit, delete, and manually trigger scheduled tasks through the API and slash-command workflow.	Use /schedule or the schedules UI/API to create a task with an agent, prompt, cron expression, and optional project path.
Scheduling and Background Services	Schedule polling loop	Polls for due tasks, computes next-run timestamps, and updates run metadata in the background.	Keep Chitty running so the background scheduler loop can monitor stored tasks.
Scheduling and Background Services	Current schedule execution status	The scheduler currently updates timestamps and logs execution activity, but the code marks full agent execution as still pending.	Use scheduling today for task management and timing metadata, but expect the actual scheduled run path to remain incomplete until the execution TODO is implemented.
Browser Extension	Connection status popup	Shows whether the extension can reach the local Chitty server.	Open the browser extension popup; it checks http://127.0.0.1:8770/health and reports connected or disconnected status.
Browser Extension	Real-browser control surface	Acts as the browser-side counterpart for Chitty browser commands so agents can operate in a real browsing session.	Install the extension in Chrome/Edge, keep the desktop app running, and use browser-enabled tasks from Chitty.
CLI and Local Services	run mode	Starts the full desktop application.	Run chitty-workspace run, or simply chitty-workspace.
CLI and Local Services	config command	Prints the current data directory and serialized application config.	Run chitty-workspace config in a terminal.
CLI and Local Services	agents command	Lists installed agents from the local database.	Run chitty-workspace agents.
CLI and Local Services	test command	Sends a headless test prompt using a selected provider and optional model.	Run chitty-workspace test "your prompt" --provider xai --model <model> after configuring a supported key.
CLI and Local Services	test-agent-builder command	Runs the agent builder flow from the command line for experimentation.	Run chitty-workspace test-agent-builder "Describe the agent you want".
CLI and Local Services	Local inference sidecar	Provides a Python sidecar process for local model inference (GGUF text, image/video/TTS generation, speech-to-text, and LoRA training).	The sidecar auto-starts when you select a local model. Install pip install -r sidecar/requirements-full.txt for full capabilities including training.

Notes on Current Implementation Status

• Scheduling is partially implemented. The application exposes schedule creation, storage, polling, timestamp updates, and manual run endpoints, but the scheduler source explicitly notes that full agent execution is still pending.
• Integrations are not yet a separate completed framework. The integrations module is currently a TODO placeholder, so most practical extension behavior today comes from native tools, marketplace packages, and connection-managed package scripts.
• Browser automation depends on the extension/browser frontend being connected. If no browser frontend is connected, browser-tool commands will fail until the extension side is available.