Pay per seat, not per answer.
Every plan ships with citations, refusals, audit replay, and the trust dashboard. You upgrade for scale, exports, and deployment posture — not for the controls themselves.
Pilot
Single team, single corpus. Get to a defended demo in two weeks.
- Up to 3 advisors, 1 reviewer
- Cited answers, threads, and refusals
- Audit log + CSV export
- PDF audit-binder export
- Hosted model routes (bring your own key)
- Role-based access + MFA
Desk
Multiple advisors and a 2nd-line reviewer working a daily queue.
- Unlimited advisors, up to 5 reviewers
- Review queue with SLAs and four-eyes sign-off
- Audit log + CSV export
- PDF audit-binder export
- Reviewer verdicts train the grounding scorer
- Role-based access + MFA
Firm
Whole firm, multi-desk, regulator-ready posture.
- Separate workspaces per desk or region
- Self-hosted: Docker + Postgres in your cloud
- Hosted or self-hosted models behind the eval gate
- Role-based access + MFA
- Record-keeping evidence: hash-chained log + PDF binder
- Tamper check on the full audit chain
No. Pricing is per seat. The default local evidence engine makes no paid model calls at all.
Yes. Admins can add an OpenRouter key, or point GlassBox at a self-hosted Ollama or vLLM model. A route is only used in production after it passes the built-in evaluation gate.
Wherever you deploy it. GlassBox runs as two containers plus Postgres, and in local evidence mode no question or document leaves your infrastructure.
Audit log → filter by date and client → export a PDF binder. Every claim, source, and reviewer override is in it.
Pricing for regulated firms always needs context. Tell us about your team and we'll reply with a quote.
Open contact form →