Jev Social is a local-first agent tool for social media research, with the core idea of letting AI decide the research steps and letting the browser provide real evidence. Users describe a research goal on Instagram, TikTok, or LinkedIn in natural language, and TypeSafe Jev selects the next bounded action, which is executed by the local socai CLI in the user's existing Chrome session. The collected evidence streams into the interface as cards and is ultimately summarized into a report with citations.
Main capabilities
- Goal routing: Jev selects Instagram, TikTok, or LinkedIn as the research platform from typed options.
- Single-step action selection: choose one step among searching, opening an observed post or profile page, reading comments, downloading a specified video, and ending the task.
- Real CLI execution: socai drives the user's existing Chrome session and returns structured evidence.
- Streaming result display: post cards remain visible, and Jev evaluates the new state and builds the final report.
- Cited reports: produce Markdown reports with source links, preserving sources, comments, media, and time information for inspection.
How to use Each step only offers actions supported by the installed socai CLI, and only for targets observed in captured results; attempted actions disappear from the next selection set. Low-confidence and malformed decisions are not executed. Login gates and step limits preserve honest partial results. By default, OpenRouter's Jev is used; the user's browser, user's session, and user's evidence all remain local. The default OpenRouter provider requires an OpenRouter key during onboarding. On macOS and Windows, the onboarding flow can install the official socai CLI if it is missing; on Linux, you must first install the latest socai CLI from source, then add it to PATH or set SOCAI_BIN. In addition, decisions can also be routed through a loopback Kev server when there is no OpenRouter key; both paths keep browser execution within the socai runtime and the user's Chrome session.
Example observations The page gives three dated local observation records: one contains 4 source link records (two searches and one post detail read selected by Jev, compiled into a Markdown report); one contains5 public result URLs (one recorded CLI search, preserving the returned post links for inspection); one shows that a video and poster were saved (capturing a 7,988,959-byte MP4 and a 95,592-byte poster, but comments were unavailable). These are single local observations, not benchmarks, and do not guarantee latency; actual site behavior, network conditions, login status, and the actions Jev selects all affect total elapsed time. v0.1.8 adds a reproducible benchmarking protocol, but live benchmark results have not yet been published.
Applicable scenarios and limitations Suitable for scenarios such as browser agent construction, creator research, and social intelligence. The project uses the MIT license and won first tool of the day on OSSDrop. OpenAI launched the Decisions API at DevDay 2026, expanding this category; however, Jev Social currently uses TypeSafe Jev through OpenRouter and has not yet integrated the OpenAI Decisions API, and will only claim compatibility after the public contract can be tested end to end.
