Jev Social

Jev Social is an open-source project from the socai-io organization. A decision model (TypeSafe Jev) picks each next bounded research operation, and the local socai CLI runs it in the user's own Chrome session. Captured posts, comments, and media stream into cards and a cited report. The latest release listed is v0.1.13.

What It Is

Jev Social is a browser-grounded social media research tool for Instagram, TikTok, and LinkedIn. The user describes a research goal, and the app routes it to a platform, then chooses one read-only operation at a time: search, open a selected profile or post, read comments, download a TikTok video when the goal explicitly asks for it, or finish. Each result goes back to Jev before the next decision. The output is a set of post cards, a table, and a Markdown evidence report with source citations.

How the Decision Loop Works

Each step offers only operations exposed by the installed socai CLI and only targets observed in captured results or given as explicit URLs. Jev does not write shell commands or arbitrary selectors. Low-confidence, malformed, or unsupported decisions do not execute, and operations already attempted are removed from the next choice set. Login gates, decision failures, and step limits produce partial results instead of a success claim. Each run records the choice, confidence, command, result summary, and elapsed time.

Reports and Evidence

After collection, an optional report step can use the OpenRouter key to synthesize a concise report from a sanitized evidence payload. Every accepted finding must cite a captured source, otherwise the app falls back to a deterministic evidence report. The project publishes recorded local runs, such as a 63.969-second Instagram run with four source-linked records, and describes them as individual observations rather than benchmarks.

Setup and Privacy

It requires Node 22+, a current socai CLI, and a decision provider: OpenRouter Jev by default, or a loopback Kev server that needs no OpenRouter key. Onboarding can install the socai CLI on macOS and Windows; on Linux it must be installed first. The README states that the app does not read the browser cookie store directly, has no analytics endpoint of its own, and sets SOCAI_TELEMETRY=0 for spawned CLI processes by default. Agent Skill installs are documented for Codex, OpenCode, and OpenClaw.