shalinda-j/vibe-research

Vibe Research is a terminal tool that runs a small team of AI agents to research a topic and produce a cited report, without you having to manage each step yourself. You give it a topic, and a planner agent breaks it into sub-questions. Several researcher agents then work in parallel, each answering one sub-question using live web search. A panel of fact-checking agents reviews the claims that come back and votes on whether each one is well supported. An editor agent then decides whether the coverage is good enough, and if it is not, the whole crew re-plans and goes back to research the gaps before a final markdown report is written with citations. While it runs, a split-screen terminal interface shows the agents working on one side and the report building on the other, and the finished report is saved to disk automatically. The tool also keeps long-term memory across runs, so later research on a related topic can build on what a previous run already learned instead of starting from nothing. The README is upfront that the tool cannot guarantee zero mistakes, since any AI model can misread a source. To manage that, every claim is tied to a real source with a link, a separate verification pass rechecks claims, and every report ends with a section describing where the tool is unsure. For anything with real consequences, such as legal, financial, or health related questions, the README recommends reading that final section and checking the sources yourself. Vibe Research can connect to over a dozen different AI backends, including Anthropic's API, Anthropic's subscription based Agent SDK, OpenAI, Google Gemini, and several other providers, plus a fully local and free option through Ollama that needs no API key at all. It also supports specialized modes, such as a medical mode that reorganizes the whole research process around clinical evidence standards like trial types and safety data, and opens every report with a note that it is not medical advice. It installs with pipx, and additional features such as subscription mode or PDF export are available as optional extras. The project is written in Python and released under the MIT license.