AI Chat Assistant

Glyph includes a built-in AI assistant that supports multiple providers (OpenAI, Anthropic, Google, Ollama, and more). Attach notes and folders as context to ground conversations in your knowledge base.

Overview

The AI panel is a resizable sidebar that provides:

• Multi-provider support: OpenAI, Anthropic, Gemini, Ollama, OpenRouter, and custom endpoints
• Context attachment: Include notes and folders in your prompts
• Conversation history: Review and restore past chats
• Two modes: Chat (conversational) and Create (single-shot generation)
• Profile management: Switch between API keys and models instantly

Getting Started

AI Providers

Glyph supports multiple AI providers out of the box:

Provider Configuration

Each provider requires:

• Name: Display name for the profile (e.g., “Work GPT-4”)
• Provider: Select from the dropdown
• Model: Choose an available model
• API Key: Your provider’s API key (stored securely)
• Base URL (optional): Custom endpoint for compatible providers
• Headers (optional): Additional HTTP headers for authentication

Chat Modes

Chat Mode

Conversational assistant with full context awareness:

• Multi-turn conversations with memory
• Follow-up questions and clarifications
• Context persists across messages
• Ideal for research, brainstorming, and exploration

Create Mode

Single-shot generation for focused tasks:

• Optimized for content creation
• Each request is independent
• Less conversational, more generative
• Ideal for drafting, summarizing, and transforming content

Switch modes in Settings → AI → Assistant Mode.

Context Attachments

Attach notes and folders to ground AI responses in your knowledge:

Adding Context

Context Types

Context Budget

Glyph automatically manages context size:

• Character budget: 100,000 characters by default
• Token estimation: Displays estimated token count before sending
• Truncation: If content exceeds budget, files are truncated in order
• Manifest: View exactly what’s included in the “Attached” section

Removing Context

Click the X on any attached item to remove it. Context persists for the current conversation until manually cleared or you start a new chat.

Using the AI Panel

Panel Layout

1. Header: Model selector, new chat button, settings, minimize
2. History: Expandable list of recent conversations
3. Chat thread: Messages from you and the AI
4. Composer: Input area with context attachments and send button

Model Selector

Click the model dropdown to:

• Switch models without losing your conversation
• View model details (context length, pricing, modalities)
• Filter models by name or ID
• Switch between API key profiles

Sending Messages

Message Actions

Conversation History

Viewing History

Click the history toggle to expand the history panel:

• Shows last 14 conversations
• Each entry displays title, model, provider, and timestamp
• Click any conversation to restore it

Restoring Conversations

Conversation Metadata

Each conversation stores:

• Job ID: Unique identifier
• Title: Auto-generated from first message
• Provider: Which AI service was used
• Model: Specific model name
• Created: Timestamp
• Message count: Number of messages exchanged
• Preview: First line of the conversation

Starting Fresh

Click the “New Chat” button to:

• Clear the current conversation
• Reset attached context
• Start with a blank slate

Profile Management

Creating Profiles

Switching Profiles

If you have multiple profiles configured:

• Click the model selector
• Use the “API keys” pill switcher at the top
• Click a profile to switch instantly
• Model list updates to show that provider’s models

Use Cases for Multiple Profiles

• Work/Personal: Separate API keys for different contexts
• Cost optimization: Use cheaper models for drafts, expensive ones for final output
• Local/Cloud: Switch between Ollama (free) and OpenAI (paid)
• Provider comparison: Test same prompt across different providers

Advanced Features

Reasoning Effort (o-series models)

For OpenAI’s o-series reasoning models:

• Configure reasoning effort: low, medium, high
• Higher effort = more thinking time = better answers (and higher cost)
• Set in profile configuration

Custom Headers

Add custom HTTP headers for authentication or routing:

[
  {"key": "X-API-Version", "value": "2024-01-01"},
  {"key": "X-Organization", "value": "your-org"}
]

Private Hosts

Enable “Allow private hosts” to connect to local AI services:

• Required for Ollama ( http://localhost:11434 )
• Required for self-hosted endpoints on private IPs
• Security warning: Only enable for trusted endpoints

Tool Timeline (Create Mode)

When using Create mode, the AI may use tools:

• create_note: Creates a new note in your space
• search_notes: Searches your knowledge base
• list_files: Lists files in a directory

The tool timeline shows:

• Which tools were called
• Tool parameters
• Results or errors
• Execution timing

Context Strategies

Minimal Context

When: Quick questions, general knowledge

Approach: Send prompts without attachments

Explain the concept of atomic notes

Single Note Context

When: Analyzing, summarizing, or transforming a specific note

Approach: Attach the target note

@daily/2024-03-10.md

Summarize the key points from this daily note

Folder Context

When: Cross-referencing multiple notes, finding patterns

Approach: Attach a relevant folder

@projects/client-work

What are the common themes across these project notes?

Mixed Context

When: Comparing specific notes against broader context

Approach: Attach multiple files and folders

@projects/q1-goals.md @projects/completed

How do my Q1 goals compare to what I actually completed?

Best Practices

1. Be specific with prompts - Clear instructions yield better results
2. Use context strategically - More context ≠ better answers; be selective
3. Review token estimates - Stay within model limits for best performance
4. Leverage history - Continue conversations instead of repeating context
5. Try different models - Each excels at different tasks
6. Use Create mode for generation - Chat mode for back-and-forth
7. Save useful responses - Click save to preserve insights in your notes
8. Name profiles clearly - “Work GPT-4” vs “Personal Claude” helps when switching

Keyboard Shortcuts

Privacy & Security

• API keys: Stored in system credential manager (Keychain on macOS, Credential Manager on Windows)
• Conversations: Stored locally in your space’s .glyph directory
• Context: Only sent when explicitly attached
• No telemetry: Glyph doesn’t log or transmit your prompts
• Provider privacy: Subject to each AI provider’s privacy policy

Common Issues

”No API key configured”

Solution: Add an API key for the selected profile in Settings → AI

”Model not available”

Solution: Check that your API key has access to the model, or select a different model

”Context too large”

Solution: Remove some attachments or use a model with a larger context window

Streaming stops early

Possible causes:

• Network interruption
• Provider rate limiting
• Model hit token limit

Solution: Click retry or reduce context size

Related Features

• Tags & Search - Find notes to attach as context
• Daily Notes - Use AI to reflect on daily notes
• Attachments - Include images in AI conversations (vision models)
• AI Setup - Configure AI providers and profiles