Skip to main content

Documentation Index

Fetch the complete documentation index at: https://mintlify.com/miu-ll/Cody-assistant/llms.txt

Use this file to discover all available pages before exploring further.

When Cody reads your recent emails from Outlook, it needs to decide which messages represent actionable tasks and which category they belong to. By default this is handled by a local keyword-based classifier, but you can optionally connect an AI provider to get significantly more accurate, context-aware classification — especially useful when email subjects are ambiguous or in mixed languages.

​
Privacy Guarantee

Before configuring AI, it is important to understand exactly what data leaves your machine:
Only the email subject line and the sender’s display name are sent to the AI provider. The email body is never transmitted.
The privacy-first design is enforced at the source in outlook.ts. The AI prompt is constructed exclusively from message.subject and message.sender fields — the bodyPreview field stays on your machine and is used only by the local classifier.
Never commit your API key to version control. Cody stores keys exclusively in %APPDATA%/cody-desktop-assistant/ on the local machine. They are never written to the project repository, never sent to Cody’s own servers, and never logged.

​
AI Providers

Cody supports two providers, selectable via the AiProvider type: 
export type AiProvider = 'azure' | 'openai'
Azure OpenAI is the recommended provider for corporate use. When your organization deploys Azure OpenAI, all API calls go to your company’s own Azure tenant. Microsoft’s enterprise data processing terms apply: the data is not used to train shared models and does not leave your corporate environment. This makes Azure OpenAI the right choice for sensitive business communications.

​
The AiConfig Interface

export interface AiConfig {
  provider: AiProvider   // 'azure' | 'openai'
  apiKey: string         // Required for both providers
  endpoint?: string      // Azure only: e.g. https://<resource>.openai.azure.com/
  deployment?: string    // Azure only: the deployment name configured in Azure AI Studio
}

​
Configuration Steps

1

Obtain credentials from your IT department

You need three values from whoever manages your Azure OpenAI resource:
  • Endpoint — in the form https://<resource-name>.openai.azure.com/
  • Deployment name — the name given to the deployed model (e.g. gpt-4o-mini)
  • API key — one of the two keys shown in the Azure portal under Keys and Endpoint
2

Open Cody Settings

Click the settings icon in the assistant panel, then navigate to the Inteligencia Artificial (AI) section.
3

Select Azure OpenAI

Choose Azure OpenAI from the provider selector. The Endpoint and Deployment fields will appear.
4

Enter your credentials

Fill in the endpoint URL, deployment name, and API key. Cody saves these fields as aiEndpoint, aiDeployment, and aiApiKey in AppSettings.
5

Trigger a sync to test

Press Sync Outlook (or use Win+Shift+S). If the AI connection is working, classified suggestions will appear with higher confidence scores (≥ 0.93) compared to local classification.
{
  "aiProvider": "azure",
  "aiEndpoint": "https://my-company.openai.azure.com/",
  "aiDeployment": "gpt-4o-mini",
  "aiApiKey": "<your-key-here>"
}

​
How Classification Works

When AI is configured and an Outlook sync runs, Cody sends a batch request containing the numbered list of email subjects and sender names. The AI responds with a JSON array assigning each email an isTask flag, a task title, a category, a subcategory, a priority, and an optional dueInDays estimate. The result is mapped back onto each InboxSuggestion:
FieldSource when AI is activeSource without AI
categoryAI response (validated against known taxonomy)Local keyword classifier
subcategoryAI response (validated against parent category)Local keyword classifier
priorityAI response (urgent / high / normal)Keyword heuristic
dueAtAI dueInDays converted to ISO timestampKeyword heuristic (today / tomorrow / +3 days)
confidence0.930.88 (keyword match) or 0.58 (fallback)
needsReviewtrue when category is "Por clasificar"true when category is "Por clasificar"

​
Fallback Behavior

No AI configured

Cody uses its built-in keyword classifier. Action verbs in the subject and body preview (e.g. “revisar”, “confirmar”, “enviar”) trigger task creation. Category assignment is regex-based and may require more manual review.

AI configured but unavailable

If the AI request fails — due to a network error, timeout (25 s), or invalid credentials — Cody falls back to the local keyword classifier silently. No sync is blocked and no error is shown to the user.

​
Reviewing Low-Confidence Suggestions

When the AI (or local classifier) assigns the category "Por clasificar" to a suggestion, the needsReview flag is set to true. Cody highlights these suggestions in the Inbox panel and prompts you to confirm or change the category before accepting the task. This keeps your task list clean without requiring you to re-sync.

​
Settings Reference

All AI settings live in the AppSettings interface: 
export interface AppSettings {
  // ...other fields
  aiProvider?: AiProvider    // 'azure' | 'openai'
  aiApiKey?: string          // API key for the selected provider
  aiEndpoint?: string        // Azure only: resource endpoint URL
  aiDeployment?: string      // Azure only: deployment name
}
Data is stored at runtime in %APPDATA%/cody-desktop-assistant/cody-data.json on the user’s machine. It is never written to the project repository or transmitted to any server other than the configured AI provider.

Build docs developers (and LLMs) love

Get started for freeTalk to us