An OSINT synthesis tool: it unifies {{ tools | length }} open-source reconnaissance tools into one provenance-tracked, deterministically prioritised brief.
Existing reconnaissance automation is good at collection and poor at judgement. Results arrive as flat, unprioritised piles: hundreds of hostnames with no indication of which matter, no record of which tool claimed what, and no way to check a surprising result short of re-running everything by hand.
Glean runs the same tools, then does the part they leave out — merging their output into one entity model, deduplicating deterministically, ranking by an explicit rubric, and keeping every finding traceable to the exact record that produced it.
Optional narration by a local LLM rewrites the prose of the top findings and nothing else. It cannot introduce a finding, reorder anything, merge two findings, or alter provenance — those are structural properties of how the brief is built, not instructions the model is asked to follow. Any finding it fails to narrate falls back to the template rather than failing the scan.
That split is the reason the output is checkable. A ranking a model produced would be a ranking nobody could audit; this one is the same rubric every time, and the tool will show you the arithmetic.
Passive and active collection are separated in code, not just in documentation. Passive sources are queried by default; anything that touches a target directly requires an explicit opt-in on every path, and no convenience default has ever been allowed to imply it.
For authorised security research only — targets you own, or are explicitly cleared to assess. The tool records the basis you state; it cannot verify it, and the responsibility stays with you.
Every significant design decision is recorded as an Architecture Decision Record in the repository — the entity schema, the adapter contract, the dedup rules, the prioritisation rubric, the brief contract, the evaluation protocol, the ground-truth method, the runner, LLM synthesis, the report view and this web interface.
They include the open questions that are still open and the mistakes found along the way, because a decision record that only lists what went well is not much of a record.