Resources

Everything we have measured, written down

Reports, docs, comparisons and engineering notes. Nothing here is behind a form or an email wall.

The benchmark report

The published benchmark results as a single downloadable document. For the reading guide — what each number means and where qbrin does not lead — start with the engineering note.

Download the PDF Read the guide

Reports

ReportBenchmark scorecardHead-to-head results with the sample size and the caveat on every row.Open the scorecard ReportSecurity benchmarkA measured SOC comparison: qbrin against RAG, an MCP agent and a naive LLM analyst, on live security data.Read the benchmark

Docs

POST /v1/DocsAPI referenceThe REST API: authenticate, ingest, ask, and get a cited answer or a clear abstain.Open the docs POST /v1/DocsAction authorization APIThe pre-execution endpoint that answers ALLOW, HOLD or ESCALATE for a proposed agent action.See the endpoint

Comparisons

Comparisonqbrin vs GleanAn architecture-level comparison, including where Glean is stronger.Read the comparison ComparisonGlean alternativesqbrin, Microsoft 365 Copilot, Coveo, Elastic and Guru: pricing models and who each fits.See the options

Engineering notes

Engineering noteHow qbrin works: evidence before answersA visual guide from connected sources to a cited answer or a deliberate abstention.Read the note Engineering noteEvidence vs verification vs authorization in AIWhy source retrieval, claim support and permission to act are three separate gates.Read the note Engineering noteqbrin benchmark results: an honest guideThe published numbers, the sample sizes, and the cases where qbrin does not lead.Read the note Engineering noteWhen an absence becomes a findingThe quietest failure in AI systems: a failed read that gets reported as a confident negative.Read the note Engineering noteThe most dangerous prompt injection is stopped by retrieval, not detectionCredential-scoped boundaries at the retrieval plane.Read the note Engineering noteA safety gate that blocks 100% of attacks and 100% of real workWhy substring grounding checks look perfect on the scoreboard and reject valid commands.Read the note Engineering noteA 400-token cap on our verifier was silently dropping correct answersHow a fixed limit turned parser errors into silent refusals.Read the note Engineering noteAI abstention: why knowing when to stop mattersThe business case and measurement discipline behind saying “not enough evidence”.Read the note

Follow the engineering blog

New measured notes as they are written. No signup needed; it is a plain blog.

Open the blog

See it answer

See it answer your hardest question.

Bring one real question your team keeps re-asking. We’ll connect a source, read-only, and show you the answer, sourced, in seconds, in a 20-minute walkthrough. Nothing changes in your tools.

One source connected, read-only
Your real question answered, with sources
Nothing changes in your tools

Pick a time

or email hello@qbrin.com