AI assist

Bring-your-own-key (BYOK) assistants for request authoring help and optional ghost-text autocomplete. Tokens never live in the versioned settings document — only hasToken presence flags are stored.

BYOK providers

Platform Availability: Full Parity (webview AI). Full provider management UI is richest on the Windows Settings overlay.
Overview: Enable OpenAI, Anthropic, Azure OpenAI, Gemini, or an OpenAI-compatible local endpoint (e.g. Ollama at http://127.0.0.1:11434/v1). Pick a default provider/model, then use the AI assist bar on requests.
How It Works (Windows App):
1. Open Settings (Ctrl+, / gear) → AI.
2. Enable AI, choose default provider, paste API token (stored via secure host APIs), set model.
3. Optionally Test connection / list models; open a request and use the AI assist bar.

Windows preferences / settings entry

How It Works (VS Code Extension):
1. Configure AI from the shared settings/Preferences surfaces available in the webview.
2. There is no contributes.configuration block — do not look for keys in VS Code Settings UI.
3. Use the same assist bar on the request editor once a provider is enabled.

VS Code preferences

Code / Configuration Examples:

{
  "ai": {
    "enabled": true,
    "defaultProvider": "openai",
    "defaultModel": "gpt-4o-mini",
    "timeoutMs": 60000,
    "maxTokens": 2048,
    "providers": {
      "openai": { "enabled": true, "model": "gpt-4o-mini", "hasToken": true },
      "anthropic": { "enabled": false, "model": "", "hasToken": false },
      "azure": {
        "enabled": false,
        "endpoint": "",
        "deployment": "",
        "apiVersion": "2024-02-15-preview",
        "hasToken": false
      },
      "gemini": { "enabled": false, "model": "", "hasToken": false },
      "openaiCompat": {
        "enabled": true,
        "name": "Local OpenAI-compatible",
        "baseUrl": "http://127.0.0.1:11434/v1",
        "model": "llama3.2",
        "hasToken": false
      }
    }
  }
}

Local Ollama-style base URL:

http://127.0.0.1:11434/v1

Edge Cases, Troubleshooting & Pro-Tips:
- defaultProvider: "none" disables outbound AI even if enabled is true.
- Azure needs endpoint + deployment + apiVersion in addition to the token.
- Tokens are not written into exported settings JSON as plaintext secrets.

GAP IDENTIFIED: The full desktop Settings overlay AI section (provider forms, token set/clear, test connection) is Windows-first. VS Code relies on in-webview AI configuration; there are no Marketplace contribution settings for keys.

Privacy controls (what gets sent)

Platform Availability: Full Parity
Overview: Explicit toggles gate whether method/path, headers, body, response body, and schema snippets are included in AI context. Defaults favor minimal disclosure (method/path on; headers/bodies off).
How It Works (Windows App): Settings → AI → privacy checkboxes; save and retry assist.
How It Works (VS Code Extension): Same privacy flags in the shared AI settings object.

Windows request builder (assist context)

VS Code request builder

Code / Configuration Examples:

{
  "ai": {
    "sendMethodAndPath": true,
    "sendHeaders": false,
    "sendBody": false,
    "sendResponseBody": false,
    "includeSchema": false
  }
}

Edge Cases, Troubleshooting & Pro-Tips:
- Never enable sendHeaders / sendBody against untrusted proxies or third-party accounts when payloads contain PII.
- Redaction still strips obvious secret-shaped values when building context — treat it as defense-in-depth, not a license to share prod tokens.
- timeoutMs (1s–300s) and maxTokens (64–32768) bound runaway calls.

Ghost-text autocomplete

Platform Availability: Full Parity
Overview: Optional editor autocomplete modes: off, debounced (default), aggressive, manual. Can use a dedicated autocompleteModel separate from the chat/default model.
How It Works (Windows App):
1. Settings → AI → Ghost-text autocomplete mode.
2. Optionally set autocomplete model.
3. Type in supported editors; accept/dismiss suggestions per mode.

Windows request tab body

How It Works (VS Code Extension):
1. Same mode enum in shared settings.
2. VS Code native IntelliSense is separate — Signoff ghost-text is inside the webview editors.
3. Use manual on slow models to avoid keystroke lag.

VS Code request tab body

Code / Configuration Examples:

{
  "ai": {
    "autocomplete": "debounced",
    "autocompleteModel": ""
  }
}

Mode	Behavior
`off`	No AI ghost text
`debounced`	Suggest after a short pause (default)
`aggressive`	Suggest more eagerly
`manual`	Only on explicit invoke

Edge Cases, Troubleshooting & Pro-Tips:
- Autocomplete still requires a configured provider with a reachable endpoint.
- For air-gapped machines, run openai-compat against a local model and disable cloud providers.
- Script editors also have non-AI qa.* snippet intel — orthogonal to BYOK.

AI assist bar on requests

Platform Availability: Full Parity
Overview: In-request assist for drafting bodies, tests, or explanations using the active provider while respecting privacy flags.
How It Works (Windows App): Open a request → use the AI assist bar → apply or discard suggestions into Body/Tests.
How It Works (VS Code Extension): Same assist bar in the webview request editor.

Windows request builder