◈ SEEKING COLLABORATORS & FUNDING — siddiki@luminarysolutions.ai
01 — OPEN DEEP RESEARCH
Research that thinks, not just retrieves.
A modular, extensible foundation for creating specialized multi-agent AI research systems. Build advanced capabilities, faster.
02 — WHY ODR-API
Stop reinventing the wheel for AI research.
Building robust AI research systems is complex. ODR-API gives you the structured, reusable foundation to focus on innovation, not infrastructure.
◇ THE CHALLENGE
Duplicated effort for search, scraping, and chunking
Complex agent coordination and fragile data flow
Difficulty adapting to new domains or tools
Scalability and maintainability bottlenecks
◈ THE SOLUTION
Modular Agencies for domain specialization
Reusable Services for core functionalities
Clear Agent orchestration patterns
Reliable data flow via Pydantic schemas
03 — CAPABILITIES
Powerful features out-of-the-box.
ODR-API provides a comprehensive toolkit for building sophisticated AI research applications.
MODULAR AGENCY DESIGN
Create specialized research domains as independent Agencies, each orchestrating their own workflow and agents for specific tasks.
REUSABLE CORE SERVICES
Leverage a shared library for search, scraping, chunking, PDF handling, and ranking across any agency.
MULTI-AGENT ORCHESTRATION
Coordinate multiple specialized LLM agents — Planner, Writer, Refiner — with customizable workflow logic.
STRUCTURED DATA FLOW
Ensure reliable data transfer and validation between agents using Pydantic V2 schemas for robust interactions.
EXTENSIBLE ARCHITECTURE
Easily add new agencies, swap services, integrate custom tools, or configure LLM providers via environment or API.
REAL-TIME STREAMING
Monitor research live with step-by-step updates streamed via WebSockets, providing full transparency.
05 — ARCHITECTURE
Solid, modular foundation.
Understand the core components that make ODR-API powerful and easy to extend.
AGENCIES
Domain-specialized orchestration units. Each Agency bundles its own logic, agents, and schemas for a specific research domain.
AGENTS
LLM-powered workers with defined roles. Coordinate within an Agency to execute sub-tasks.
SERVICES
Shared, non-LLM capabilities called programmatically by Agencies and Agents.
SCHEMAS
Pydantic V2 models defining data contracts between all components — the glue of the system.
04 — HOW IT WORKS
Orchestrated agent pipeline.
See the coordinated flow of agents and services inside the deep_research agency.
DEFINE & PLAN
Provide a research query. The Planner agent generates a structured plan with search queries and a writing outline.
Input: User Query → Output: WritingPlan, SearchTasks
GATHER & PROCESS
The system executes searches, scrapes web content via Crawl4AI, handles PDFs, chunks text, and ranks for relevance.
Services: Search, Scraper, Chunking, Ranking
SYNTHESIZE & WRITE
The Writer agent drafts the report using processed context, structuring it according to the plan with citations.
Input: Processed Content, Plan → Output: Draft Report
REFINE & ITERATE
The Refiner agent iteratively enhances the draft, incorporating only new information from further searches.
Loop: Search → Process → Refine → Output: Refined Report
FINALIZE & DELIVER
The system assembles the final report, formats citations, adds references, and delivers via WebSocket stream.
Output: Final Report, Usage Stats
06 — TECH STACK
Modern, robust tooling.
Leveraging powerful libraries and services for optimal performance and developer experience.
BACKEND FRAMEWORK
AI & CORE SERVICES
FRONTEND DEMO