GLChat DeepResearch Profiles

GLChat DeepResearch offers three user-facing profiles that balance research depth, cost, and speed. This document is organized into two sections:

1. GLChat Deep Research Profiles Overview - Conceptual explanation of each profile

2. Current Implementation - Technical implementation details and backend providers

Deep Research Profiles Overview

GLChat provides three distinct research profiles to match different user needs:

Essentials Profile

Essentials is the cost-effective research profile designed for quick insights and high-volume use cases.

Profile Characteristics:

• Cost: Low (most cost-effective)

• Speed: Fast (2-10 minutes typical)

• Depth: Moderate — Focuses on key points and core insights

• Best For: Quick decision-making, time-sensitive queries, budget-conscious applications

Use Cases:

• Quick decision-making support

• Fast orientation on new topics

• Time-sensitive queries

• High-volume research tasks

• Budget-conscious applications

Comprehensive Profile

Comprehensive is the thorough research profile that delivers professional-grade, in-depth research results.

Profile Characteristics:

• Cost: High (most expensive)

• Speed: Slower (10-30+ minutes depending on complexity)

• Depth: Maximum — Covers all relevant angles, details, and supporting data

• Best For: Professional research, academic work, complex queries requiring multiple sources

Use Cases:

• Professional research tasks

• Academic or scientific research

• Situations where precision matters more than speed

• Complex queries requiring multiple sources

• When comprehensive coverage is critical

Internal Profile

Internal is a specialized research profile that searches both web and internal organizational data (e.g., Google Drive).

Profile Characteristics:

• Cost: Low (similar to Essentials)

• Speed: Fast to Moderate (2-10 minutes typical)

• Depth: Moderate — Similar to Essentials, but with access to internal data sources

• Best For: Organizational knowledge management, internal document search, compliance queries

Use Cases:

• Searching company documents and files

• Finding information in Google Drive

• Combining internal and external knowledge

• Organizational knowledge management

• Compliance and audit queries requiring internal data

Current Implementation

Key principle: One profile = One provider per query.

Each profile is mapped to a specific backend provider optimized for that use case.

Essentials Profile Implementation

Backend: OpenAI Deep Research with o4-mini-deep-research model

Technical Details:

• Model: o4-mini-deep-research

• Tools: Web search, code interpreter

• Native Tools: Uses OpenAI's native tool-calling capabilities

Required Environment Variables:

BACKEND_FEATURE_ESSENTIALS_DEEP_RESEARCH_ENABLED=true
OPENAI_API_KEY=<your-api-key>

Status: ✅ Production Ready

Comprehensive Profile Implementation

Backend: Parallel AI Deep Research with ultra8x processor

Technical Details:

• Provider: Parallel.ai

• Processor: ultra8x (high-performance processor)

• Processing Time: 5-10+ minutes for ultra processor

• Tools: Multiple specialized research tools and processors

Fallback Handling: When Parallel AI quota is exceeded (HTTP 402 or 429):

1. QuotaExceededError is caught by TryCatchStep

2. Warning event (cdr_to_edr) is emitted to notify the user

3. Automatically falls back to Essentials profile

4. Request completes without user intervention

Required Environment Variables:

BACKEND_FEATURE_COMPREHENSIVE_DEEP_RESEARCH_ENABLED=true
PARALLEL_AI_API_KEY=<your-api-key>
PARALLEL_AI_PROCESSOR=ultra8x

Status: ✅ Production Ready

Internal Profile Implementation

Backend: OpenAI Deep Research with Google Drive connector integration

Technical Details:

• Model: o4-mini-deep-research (planned)

• Tools: Web search, code interpreter, Google Drive connector

• Integration: Google Drive via OAuth tokens

Google Drive Integration Flow:

1. Token Generation: Refresh token → Access token via Google OAuth

2. Caching: Access token cached in Redis with TTL matching expiration

3. Encryption: Token encrypted using GLChat encryptor before storage

4. Connector Injection: Access token passed to OpenAI's native connector

5. Research: OpenAI can search and access Google Drive files during research

Required Environment Variables:

BACKEND_FEATURE_ESSENTIALS_DEEP_RESEARCH_ENABLED=true
OPENAI_API_KEY=<your-api-key>
OPENAI_DEEP_RESEARCH_GOOGLE_REFRESH_TOKEN=<refresh_token>
OPENAI_DEEP_RESEARCH_GOOGLE_CLIENT_ID=<client_id>
OPENAI_DEEP_RESEARCH_GOOGLE_CLIENT_SECRET=<client_secret>

Status: ✅ Production Ready