Export findings to Supabase and FalkorDB
Keep findings and verdicts in a store you run: a table you can query in Supabase, or an evidence graph you can traverse in FalkorDB. Nothing is written until you press Send, and each write goes only to the address you configured.
Before you start
Section titled “Before you start”- A result in Keryn Web with the findings (and, if you want them, verdicts).
- For Supabase: a project you own and a key that may insert into one table. For FalkorDB: either a relay in front of your database, or a client to load a script yourself.
- Findings describe weaknesses in your code. Decide who may read the destination before you create it.
Supabase: findings as rows
Section titled “Supabase: findings as rows”- In Settings, Connectors, choose Supabase. The app shows the SQL that creates the table. Run it in the Supabase SQL editor. The table has row level security on and grants nothing to
anon, because a project’s anon key is public. - Enter the project address, the table name and a key. The
service_rolekey bypasses row level security, so use it only in a tab you trust and do not choose “remember for this tab” for it. - Press Test. It checks the address and key without sending a finding.
- Press Send. The panel names the host and the number of rows first; confirm to write.
The columns, the key header and the exact fields are on the Supabase page, generated from the product’s catalog. A verdict, its confidence, grounds and fix are separate triage_* columns; the proof level is its own pair of columns (proof_level, proof_rung) and a verdict never changes it.
Query it the way you query anything in Postgres. The table is keryn_findings by default, with one row per finding, so “the errors that reached proof level 3 or higher” is a where clause on severity and proof_level.
FalkorDB: the evidence graph
Section titled “FalkorDB: the evidence graph”The graph is the evidence graph of one scan, written as openCypher statements: a Scan node that REPORTED each Finding, which is IN_FILE a File, VIOLATES a Rule and is TAGGED with a Taxon (a CWE or OWASP entry), plus the Component nodes of the AI-BOM that the scan INVENTORIED and the code USES. There are two routes.
- Through a relay. Deploy the Keryn relay on Vercel or Netlify with the
falkordbpackage, point Keryn Web at it, press Test (the relay’s health route says whether graph writes are enabled) and Send. The database credentials stay in the relay’s environment, never in the tab. - Without a relay. Download the graph as an openCypher script and load it yourself with
redis-cli, FalkorDB Browser or any openCypher client. Nothing leaves the page, because you carry the file.
The labels and relationships are on the FalkorDB page.
Keep the keys small
Section titled “Keep the keys small”Secret fields stay in the tab’s memory. Use a key that can write one table or one graph and nothing else, rotate it when the tab is shared, and use Forget on the connector when you finish. Every request is pinned to the one origin you entered, with no cookies and no referrer; a redirect to another origin is refused.
What you just proved
Section titled “What you just proved”You moved evidence from a tab to a store by choice, across one edge that you configured and confirmed. The page cannot write anywhere else: the content security policy of the page lists the hosts a connector may reach, and the connectors overview prints that list.