Repository navigation
Replaces the interactive shell with a web UI served on 127.0.0.1, built on a new side-effect-free service layer (src/osint_service.py) that both the app and the LLM tools share.
Two ways to drive it: AI mode, where a local model via Ollama picks which lookups to run and summarizes the answer, and Base mode, where all 28 commands are listed and you set their parameters yourself. No external LLM key is ever needed.
New analyses: "About this account" (country of registration, creation date, username changes), posting-times heatmap, profile comparison on followers, followings or both, story highlights, suggested profiles, hashtag and place search that need no target, and a profile lookup that surfaces every field Instagram exposes, bio links included.
Cost control, since the backend is the part with a real budget: per-request cache shared across commands and targets, a request estimate before running, remaining HikerAPI credit in the header, a hard per-run cap, server-side cancellation, and retry with backoff on transient failures.
Saving and sharing: dossiers that can be reopened without spending a request and diffed against a later run, JSON/CSV export, a media zip, and a standalone HTML report.
The HikerAPI key can be pasted into the UI: verified before it is accepted, never readable back, and reachable only from localhost.
Removed the shell and its five dependencies; one of them (pyreadline 2.1) was broken on Python 3.10+, so installing used to fail outright on modern Windows. Docker now serves the web app as a non-root user on 127.0.0.1, with config, cache and dossiers mounted from the host. CI actually runs the 125-test suite, which needs no API key, quota or network.