Documentation

Overview

Aegis AI Code Editor is a Tauri + React desktop shell: you open a project folder, edit in Monaco, chat in Omni-Chat (OpenRouter-backed models), run a real PTY terminal, index code into the Knowledge Vault, and use optional MCP tools. Billing and credit packs are handled through Stripe inside the app UI.

All primary routes (Omni-Chat, Vault, Analytics, Billing, Settings, and more) live in the left navigation and the application menu. Use Help → Welcome for shortcuts and Help → Download & Updates for installers.

Omni-Chat

Omni-Chat is the streaming assistant panel. Attachments, model selection, and conversation history are stored in the local app database. Long-running agent tasks can emit tool calls (file writes, MCP, etc.) according to your policy.

Voice and hygiene

Optional voice dictation and readable markdown output keep long sessions manageable. Fenced code blocks can be copied or run where policy allows.

Monaco editor

The editor uses the same Monaco stack as VS Code–family tools: multi-tab editing, syntax highlighting, large-file friendliness, and workspace-relative paths. Inline AI completions are available when enabled in Settings.

Project scope

Open a folder from File → Open Folder. Autosave, formatting, and integrated diff flows follow the workspace you have loaded.

Knowledge Vault (RAG)

The Vault indexes selected projects into a local vector store (SQLite-backed) for semantic retrieval. You can reindex, delete per-project memory, and tune how retrieved chunks accompany chat prompts.

Privacy

Embeddings and chunk text stay on the machine unless you configure cloud models for embedding or chat through OpenRouter—those calls follow your API keys and network policy.

Agents & orchestration

The autonomous agent loop plans work, edits files, and can execute shell steps under an allowlisted EXEC policy. Orchestration lets a main agent coordinate staged sub-agents so risky changes stay reviewable before landing on your live tree.

Rules

Agent Rules capture project-specific instructions the loop must respect—coding standards, forbidden paths, and tone for generated commits or messages.

Git

The Git page surfaces status, history, diffs, commits, and safe automation helpers driven by the local git CLI. Review AI-generated patches before you commit.

Terminal

The integrated terminal is a real PTY session (ConPTY on Windows) bound to your workspace directory. Multiple tabs, show/hide, and standalone terminal routes mirror a classic IDE layout.

Search

Workspace text search uses ripgrep-style scanning across the open project with filters for paths and file types. Use it before you refactor or when tracing symbols across large trees.

MCP (Model Context Protocol)

Connect MCP servers over stdio JSON-RPC to expose tools to the assistant. Tool execution respects your workspace boundaries; configure servers in Settings and validate trust before enabling high-privilege tools.

Billing & credits

The Billing area shows credit balance, Stripe checkout for packs, and ledger-style history. Keys and customer identifiers are handled through secure flows—never commit secrets into a repository opened inside the IDE.

Analytics

Analytics summarizes agent task history, model usage, and related signals so you can audit spend and outcomes. Data is stored locally with the desktop app, not in the marketing site.

Settings & security

Settings centralize API keys (stored in the OS credential store where supported), EXEC policy, MCP configuration, editor preferences, and security guidance. Review the in-app security callouts and your organization’s policies before enabling automation.

Updates

The desktop app can check for signed updates when configured. Installers for Windows, macOS, and Linux are published on GitHub Releases; the marketing home page also links a CDN-hosted Windows installer when available. Always prefer HTTPS download URLs from sources you trust.