Pokee AI announced the release of PokeeResearch-7B — an open 7-billion AI developed for deep research on the web. The model can search for information online, gather evidence, and verify its own responses. It is available with all code, weights, and a research paper describing the reinforcement approach and multi-level logic.
PokeeResearch-7B is suitable for engineers, analysts, startups, and research teams. Users can run it locally or through vLLM. The model integrates web search, reads content, creates parallel research threads, independently verifies results, and recovers from tool errors.
The repository includes documentation for testing on A100 80GB, and options for running via CLI and Gradio are available. All materials are openly accessible on Hugging Face. Pokee AI notes that the license is Apache-2.0, but the article mentions MIT, so the licensing issue is still being clarified.
Pokee AI also presents a preview of the hosting Deep Research API, which is positioned as more cost-effective compared to other solutions. The company targets its product at those who create reports with reliable references and require dependable source handling.
