omni-research

What problem does it solve?

Performing reliable web research with verifiable citations is time-consuming and error-prone when done in a single prompt. Omni-Research runs a multi-cycle autonomous loop that scouts sources, verifies claims by reading pages, synthesizes findings, and accumulates a growing knowledge base with traceable evidence so users get reproducible, source-backed research rather than one-shot, hallucinated answers.

Core Features & Use Cases

• Interactive setup and configuration: Guided prompts collect topic, type (product/marketing/production/research), seed sources, research lines, and scope before launching an autonomous run.
• Multi-phase autonomous loop: Phase 0 source scouting, Phase 1 seed/source mining, Phase 2 hypothesis-driven cycles with periodic discovery, scoring, and gate decisions (PROCEED/REFINE/PIVOT).
• File-driven orchestration and steering: Generates program.md, research.md, experiments.tsv, steer.md and runs a background agent that checks steer.md for mid-run redirects; supports recovery and BRIEF.md generation from saved outputs.
• Strict source verification and inline citations: Requires WebFetch confirmation for every cited claim, scores sources for authority/recency/relevance, and flags unverified or conflicting evidence.
• Use cases: product-market research, go-to-market planning, technical literature reviews, competitive analysis, and automated briefing generation for stakeholders.

Quick Start

Start a background research run for the topic "impact of AI on elder care" by invoking /omni-research and answering the setup prompts.