FrameworkPicker
There are 547 distinct open-source AI agent frameworks in the LocalAIAgents dataset, and most "best framework" lists are vibes. Answer four questions about your language, the job, your license needs and how alive the project must be, and the Picker ranks the dataset itself: a top 3 with the recorded stars, license and last-commit date on every card.
Loading the LocalAIAgents dataset...
How to read this: the shortlist is a starting point, ranked mechanically from the dataset's recorded fields. Check the framework's own repository before you commit to it; stars and commit dates move, and a framework that fits your answers still has to fit your team.
How the ranking works
- Hard filters first. Your language and job answers are never relaxed: a framework must be recorded in your language (or either language, if you picked Either) and carry at least one capability tag that matches your job.
- Then a deterministic score. Matches are ranked by how many of the framework's recorded capability tags match your job, then by GitHub stars as recorded, then by forks, then by name. No adoption score is used: the dataset has one, but the public feed behind this tool does not expose it, so the Picker ranks on what it can show you.
- Relaxation, stated out loud. If fewer than 3 frameworks match every answer, the Picker relaxes your license preference first (one notch at a time), then your actively-maintained preference, and says on each card what was relaxed. It never pads the top 3 with frameworks that fail your language or job.
- Job tags. A capability tag matches a job when one of its hyphen segments or a published substring matches: chatbot work looks for segments like "chatbot", "assistant" and "chat"; multi-agent work for "multi-agent" and "swarm"; workflow automation for "workflow", "automation", "no-code" and "low-code"; coding for "coding", "developer-tools" and "claude-code"; research and data work for "research", "rag" and "data-analysis". The full rule set is in the build report.
- Momentum. The dataset records a last-commit date per framework. "Actively maintained" means that date is on or after the cutoff, six months before the newest commit in the dataset (...).
Where the data comes from
The Picker is generated from the same agent-framework dataset that powers LocalAIAgents: ... frameworks with recorded GitHub stars, forks, primary language, license and capability tags. Fields the dataset does not record for a framework are shown as not recorded on its card, never filled in by guesswork. The site table lists 548 rows; one of them (AutoGPT.js) is an exact duplicate entry, so the Picker works over the 547 distinct frameworks. The dataset feed also carries no license value for 56 frameworks (the underlying table classifies their licenses as "other"), so those show as license not recorded and only match a no-preference license answer. Projects change hands and licenses get re-read, so confirm the current state on the framework's own repository before you rely on a result.