Deep Dog 2
An open-source deep-research agent that supports dispatching specialized sub-agents across multiple platforms. It features a minimal installation process and cache-friendly execution, and can connect to multiple large-model service providers and search engine APIs.
Timeline (last 90 days)
Deep Dog 2 ranked 5th overall on the DeepResearch Bench evaluation, making it the best-performing open-source solution among those tested.
Deep Dog 2 is an open-source Deep Research Agent.
Deep Dog 2 can be installed from its GitHub repository with a single pip command.
Deep Dog 2 is compatible with both Exa and Tavily search engines.
Deep Dog 2's default configuration uses the DeepSeek V4 Flash model for the supervisor agent, research sub-agents, and the final drafting stage, with Exa used for retrieval.
Deep Dog 2 has built-in platform-specific specialized sub-agents, including Web, PubMed, Reddit, Substack, SEC Edgar, and Arxiv.
Deep Dog 2's research results are returned as Markdown strings.
An architecture capable of dispatching domain-specific sub-agents has a structural advantage over general web search solutions in multi-source retrieval evaluations.
Deep Dog 2 is fully free and open-source under the MIT license.
The author of Deep Dog 2 has explicitly stated that they will not commercialize it or offer a hosted service, and the only potential cost comes from the APIs of the service providers used.
1 more timeline events
All Facts (10)
Deep Dog 2 is an open-source Deep Research Agent.
50%UnverifiedDeep Dog 2 can be installed from its GitHub repository with a single pip command.
50%UnverifiedDeep Dog 2 is compatible with both Exa and Tavily search engines.
50%UnverifiedDeep Dog 2's default configuration uses the DeepSeek V4 Flash model for the supervisor agent, research sub-agents, and the final drafting stage, with Exa used for retrieval.
50%UnverifiedDeep Dog 2 has built-in platform-specific specialized sub-agents, including Web, PubMed, Reddit, Substack, SEC Edgar, and Arxiv.
50%UnverifiedDeep Dog 2's research results are returned as Markdown strings.
50%UnverifiedAn architecture capable of dispatching domain-specific sub-agents has a structural advantage over general web search solutions in multi-source retrieval evaluations.
50%UnverifiedDeep Dog 2 is fully free and open-source under the MIT license.
50%UnverifiedThe author of Deep Dog 2 has explicitly stated that they will not commercialize it or offer a hosted service, and the only potential cost comes from the APIs of the service providers used.
50%UnverifiedDeep Dog 2 provides a pre-packaged integration layer that allows developers to trigger the complete multi-agent research pipeline simply by calling run_research(), without needing to rebuild the orchestration system from scratch.
50%