AI assistant

HarborClient includes a built-in AI sidebar for working with your API requests. Chat with models from OpenAI, Claude, or Google Gemini using your own API keys, or use models provided by a connected Team Hub. When a hub offers a model, HarborClient prefers the hub and never sends you the server-side provider keys.

API keys are encrypted and stored locally on your machine. HarborClient uses the OS keychain when available; on systems without Secret Service support it falls back to a local encryption key in your application data directory. On Linux, OS-backed encryption typically requires GNOME Keyring or KWallet to be running.

API keys

Before you can chat with personal models, add at least one provider API key:

1. Open File → Settings (or Cmd/Ctrl+,).
2. Select the AI section in the settings sidebar.
3. Enter one or more keys:
   • OpenAI API key
   • Claude API key
   • Google Gemini API key
4. Click Save.

If you open the AI sidebar before any keys or Team Hub models are available, click Open AI settings in the prompt to jump straight to the AI section.

Only available models appear in the chat model picker. Labels show whether a model uses Team Hub or Personal keys. Supported models:

Model	Provider
GPT-4o	OpenAI
GPT-4o Mini	OpenAI
Claude 3.5 Sonnet	Claude
Claude 3.5 Haiku	Claude
Gemini 1.5 Pro	Google Gemini
Gemini 1.5 Flash	Google Gemini

See Settings → AI for key storage details.

Opening the AI sidebar

Show or hide the AI panel from:

• View → AI in the menu bar
• The robot button in the footer (Show/Hide AI sidebar)

Your sidebar visibility preference persists across app restarts. The AI sidebar is hidden while Settings, Team Hub, or Sharing Keys are open.

Using chat

The AI sidebar provides a tabbed chat panel. Use the controls in the chat tab bar and composer to manage conversations.

Action	How
New chat	+ button in the chat tab bar
Open previous chat	History button (clock icon) → select a chat
Close tab	Close button on the tab (the chat remains in history until deleted)
Delete chat	History menu → Delete
Choose model	Model dropdown in the composer (per chat)
Send message	Send or Enter (Shift+Enter for a newline)

Chats and messages are stored locally in the local database. New chats start as New Chat; the tab title updates from your first message. Open tabs and the active selection persist across restarts. While the model is working, the UI shows Thinking…; errors appear inline below the composer.

Inline /ask in script editors

While editing JavaScript in a pre- or post-request script, type /ask for quick AI help without opening the sidebar. The model replaces the entire /ask line with a short answer — either // comment lines or executable JavaScript — so the command is not left in the script at send time.

Where it works

• Expanded inline script editors in request PreRequest and PostRequest tabs
• The same inline editors in collection settings PreRequest and PostRequest tabs
• Requires AI to be configured — a personal API key or a connected Team Hub with LLM models (same gate as the Ask AI wand button)
• Not available on snippet-linked rows, which show read-only snippet source; use Ask AI about a script on those rows instead

How to use

Input	Behavior
/ask on its own line	Opens a small anchored dialog. Type a short question, choose a model, then Send or Enter (Shift+Enter for a newline). Stop cancels an in-flight request. Escape or click outside dismisses the dialog and removes the /ask line.
/ask followed by a question on the same line	Runs immediately. The line shows Thinking… until the answer is applied.

You can also pick /ask from the editor slash-command menu when AI is available.

What you get back

• Explanations — one or more // comment lines inserted in place of the /ask line
• Code or examples — JavaScript that replaces the /ask line, using the hc API (not Postman pm syntax)
• Errors — shown inline in the dialog, or as a toast when you used inline /ask question mode

Model selection

Script /ask requests use a separate model preference from sidebar chats. HarborClient picks the last model you used for a successful /ask request, then falls back to the active sidebar chat model, then the first available model. The modal includes a model dropdown when you need to override the default.

Examples

Pre-request script:

/ask How do I set a request header?

Post-request script:

/ask Write a test that asserts status 200

Inline /ask vs sidebar Ask AI

	Inline /ask	Ask AI wand / @ token
UI	In-editor dialog or one-shot	AI sidebar chat
Scope	Replace one line or a short answer	Multi-turn; can edit the whole script row
Snippet-linked rows	Not available	Use this

For broader script edits and multi-turn conversations, use the Ask AI wand or an @<request-id>.<pre\|post>.<index> reference in the sidebar — see Script references below.

What the assistant can do

The assistant reads live HarborClient state through built-in tools. It inspects your workspace and responses, and can perform a few actions when you ask.

Inspect workspace

• The collection selected in the sidebar, and all collections (with variables, headers, auth, and scripts)
• Saved requests in a collection
• Environments and which one is active
• The request open in the editor (summary and full draft, including cookies for the URL host)
• The saved request highlighted in the sidebar

Inspect responses

• Last response summary (status, headers, short body preview, test results)
• Full response body (capped)
• JMESPath queries against JSON response bodies (for example length(data.items), data.users[*].id)

Actions (only when you ask)

• Send the active request (equivalent to clicking Send)
• Switch or clear the active environment
• Modify the active request (method, URL, params, headers, body, auth, pre/post scripts, cookies)
• Update a specific pre- or post-request script on the active request when your message includes an @<request-id>.<pre|post>.<index> reference (for example @42.post.1)

Changes from the assistant appear in the editor immediately. Saved requests show as unsaved until you save them yourself.

Constraints

• The assistant uses tools to read live app state — it should not invent URLs, headers, or test results.
• It will not send requests or change environments unless you explicitly ask.
• Request edits update the open tab draft only; they are not saved to the collection automatically.
• Very long conversations may hit model context limits; start a new chat if that happens.

Example prompts

• “What collections do I have?”
• “Summarize the last response for this request.”
• “How many items are in the response array?”
• “Send this request and tell me if the tests passed.”
• “Add a post-request script to the script list that asserts the response status is 200.”
• “@42.post.1 — add a test that asserts the response status is 200.”
• “Using @active.pre.1, set the Authorization header from the token variable.”

Script references (@request.stage.index)

When a message starts with @<request-id>.<pre|post>.<index> (for example @42.pre.1), the assistant treats it as a pointer to a specific script row in the pre-request or post-request stage on the request open in the editor. It reads that script's source, then applies your requested changes to the same row. You can type the reference yourself or use the Ask AI button on a script row to open a new chat with the token prefilled. See Request scripts — Ask AI about a script for the button workflow and token format.

For quick line-level help while typing in an inline script editor, use Inline /ask in script editors instead — it stays in the editor and does not open the sidebar.

What's next

• Settings — API key storage and other application preferences
• Cookies — cookie jar storage and File → Cookies management
• Making requests — the request editor the assistant reads
• Environments — variables the assistant can inspect and switch
• Collections — saved requests the assistant can list