Chapter 8 — AI Assistant and Sidebar

AidaIDE has two distinct ways to use AI: a sidebar that lives next to your code for casual back-and-forth, and a set of AI feature tabs for managing more complex multi-agent workflows. This chapter covers the everyday AI tools. The visual canvas for building agent pipelines is in Chapter 9.

AI sidebar — dockable panel on the left. Chat with the model, ask about the file you have open, get inline edit suggestions. One conversation at a time.
AI feature tabs — full notebook tabs for purpose-built workflows: Claude Agent Manager (model setup), AI Workflows (recipe runner), AI Agent Portal / Persona Portal, Live Coding Agents, Agent Hierarchy Builder.

You can have both open. The sidebar is for "quick question," the tabs are for "this is its own task."

Toggle the sidebar with AI → Toggle AI Sidebar (Ctrl+Shift+I).

AI Chat (`Ctrl+Shift+A`)

The simplest entry point. Opens the AI sidebar focused, with the chat composer at the bottom.

Things it can do out of the box:

Answer general questions about your code if you have a file open and Include current file is ticked.
Edit the file on request — pops up a diff preview before applying.
Run shell commands on the focused connection if you grant permission per command.
Reference the file tree of your project if Include workspace context is on.

The chat is stateful within a session. Sessions live until you close them or hit New Chat.

Claude Agent Manager (`Ctrl+Shift+M`)

Open from AI → Claude Agent Manager or the green ★ AI Assistant toolbar button.

The control panel for AI providers and agent presets:

Providers — Anthropic Claude (recommended), OpenAI, Azure OpenAI, local Ollama. Each has its own API key field; keys are stored in the vault.
Model picker — Claude Sonnet 4.6 / Opus 4.6 / Haiku 4.5 (or your provider's equivalents). Per-agent override possible.
Agents — named profiles with custom system prompts. The default "Pair Programmer" agent ships ready to use. Create your own for "Embedded mentor," "Security reviewer," etc.
Test connection — sends a one-token ping to verify the key.
Usage — token counters per agent, per day. Helpful for cost-tracking.

You must have at least one provider configured for chat to work. The Onboarding Wizard prompts for this on first launch.

Toggle Inline Suggestions

AI → Toggle Inline Suggestions.

When on, the editor shows greyed-out completion text after your cursor. Press Tab to accept, Esc to dismiss, keep typing to override. Off by default — turn it on once you have an API key configured and a model that is fast enough not to be annoying (Haiku is recommended for this use).

If suggestions are slow or wrong, narrow them with Settings → AI → Inline Suggestions → Context window.

AI Workflows

Open from AI → AI Workflows.

A recipe runner for AI tasks that touch your filesystem or connections. Each workflow is a YAML/JSON file describing:

Inputs (files, paths, parameters).
A series of AI calls (with prompts).
File-system and shell side effects.
Output validation.

Built-in workflows include:

Refactor file — propose, diff, apply.
Generate docstrings — across a project.
Convert script to systemd unit — for the focused connection.
Triage logs — feed journalctl output, get a ranked list of issues.
Translate API client — Python to Go, Node to Python, etc.

Click New Workflow to author your own. Saves to ~/.aidaide/workflows/.

AI Agent Portal

Open from AI → AI Agent Portal.

A chat-style UI where you talk to a single agent (one of the ones you defined in Claude Agent Manager). Differences from the AI sidebar:

Each agent has its own persistent history.
Agents can have tools attached (shell, file write, web fetch). Tool calls show inline.
Useful for long-running "I have an assistant who knows this project" workflows.

Persona Portal

Open from AI → Persona Portal.

Similar to AI Agent Portal but oriented around personas — pre-shaped characters with personality, tone, and lane. Ships with personas like "Wes, the patient mentor," "Macy, the interactive guide," "Astrid, the security skeptic." You can create your own.

Personas are agents plus narrative styling. Useful if you want the model to consistently sound a certain way (e.g. for documentation generation, for student-facing tutoring).

Live Coding Agents (`Ctrl+Shift+J`)

Open from AI → Live Coding Agents.

An always-on "pair programmer over your shoulder" mode. Differences from chat:

The agent watches what you do (file opens, edits, runs) and offers suggestions proactively.
Suggestions appear as a non-modal toast in the bottom right. Click to expand, accept or dismiss.
You can pin instructions ("I am refactoring auth.py, ask before touching anything else").

Token usage is higher than chat. Live Coding is best for focused sessions, not all-day idle.

Saving and loading sessions

AI → Save Chat Session (Ctrl+Shift+Alt+S) — writes the current sidebar conversation to disk. Pick a name.
AI → Load Chat Session (Ctrl+Alt+L) — pick from saved sessions.

Saved sessions are JSON files in ~/.aidaide/chat_sessions/. They include the model used, system prompt, every turn, and any attached files. You can share them with teammates or restore an interrupted thread.

License gating

Some AI features are gated behind paid tiers:

AI Hierarchy plan unlocks the Hierarchy Builder and the multi-agent runtime.
AI Agents (basic) covers Claude Agent Manager, AI sidebar, AI Chat, Inline Suggestions, Live Coding Agents, AI Workflows.

If a feature is locked, the menu item is visible but greyed out, and clicking it pops a "Enter License Key" link → Chapter 17.

Tips

Reference a file — paste its path with @ in the chat composer (@src/main.py). The model sees the whole file in context.
Reference a line range — @src/main.py:42-58.
Stop a response — press Esc while it streams.
Branch a conversation — right-click any message → Branch from here. Useful when an answer was almost right; rewind and try a different prompt.

Terms & Setup

API key

A long random string from your AI provider. AidaIDE stores it in the vault.

Anthropic Claude: sign up at console.anthropic.com, create a key under API Keys. Paste into Claude Agent Manager → Anthropic → API Key.
OpenAI: platform.openai.com → API Keys → Create. Same flow.
Azure OpenAI: keys come from the Azure Portal; you also need the endpoint URL and the deployment name. All three go into the Azure provider config.
Ollama (local): install Ollama on your machine. Point AidaIDE at http://localhost:11434. No API key required.

"I don't have an API key yet" — setup paths

You can exercise the AI sidebar, AI Chat, AI Workflows, and the Hierarchy Builder before you pay any provider a cent. Pick the path that fits your situation:

Free local models via Ollama (recommended for tinkering). Install Ollama from ollama.com/download (Win/macOS/Linux). On first launch it listens at http://localhost:11434. Pull a small model: ollama pull llama3.2 or ollama pull qwen2.5-coder:7b. In Claude Agent Manager → Providers, pick Ollama, leave the API key blank, set the host to http://localhost:11434, hit Test Connection — you should see a green tick. No keys, no spend, fully offline. Slower than cloud for big prompts but fine for editor inline suggestions and chat. See Chapter 8 §Claude Agent Manager.
Anthropic Claude starting credits. Sign up at console.anthropic.com — new accounts receive a small starting credit (currently $5, may change) sufficient for hours of Haiku or Sonnet use. Create a key under Settings → API Keys, paste into Claude Agent Manager → Anthropic. The vault encrypts it at rest.
OpenAI starter credit. platform.openai.com — same flow, occasional new-account credits. Lower-cost models (gpt-4o-mini) make a s