The questions we actually get.

No. It drafts an answer for each extracted requirement and exports those answers plus the compliance matrix. Narrative sections (executive summaries, win themes, past-performance volumes) are still written by your team.

Scanned pages are OCR’d. The page is marked OCR with lower confidence, and a citation still points to the page it came from. Check the OCR pages the app marks before you rely on them.

No. Typezo runs as a dedicated single-tenant instance that we operate. There is no self-hosted install.

No. It exports the compliance matrix (XLSX) and the draft (DOCX). Submission to SAM.gov, PIEE or any agency portal stays with you.

You do not have to trust the prose. Every citation quote is checked against the stored page text; the page and offsets come from the parser, not the model. Every number, date and dollar figure in an answer must appear in one of that answer’s cited quotes. A run that fails verification becomes needs_review and cannot be exported as clean.

Verification checks grounding, not judgement. A cited answer can still be weak, and a person decides whether it belongs in the bid.

The run is marked needs_review and the export is watermarked. There is no clean export of an unverified run. A run that fails verification cannot be exported as clean.

A gap is different. A requirement with no supporting span is a result, not a verification failure: it stays missing, the matrix gets a gap row, and the run can still pass.

No. A help answer is drafted only from published article spans, and the same deterministic verifier re-checks quotes, pages and figures. If nothing in the knowledge base supports the question, there is no answer: it opens a ticket in the tenant-wide queue instead.

Documents are parsed and stored on a dedicated instance for your workspace. Prompts pass through the model gateway to a zero-retention provider that is contractually barred from training on or retaining them. The control plane forwards prompt bytes in memory but does not write request or response bodies to disk, does not store or log document text, and keeps no document text in its database. Customer documents are never used for training, by Typezo or by a provider. The honest version is “forwarded, not stored at the control layer”, not “never seen”.

No. Typezo does not hold FedRAMP, CMMC or SOC 2. If a questionnaire requires one of those today, this is not the right vendor, and this page will not imply otherwise.

One tenant runs on one instance, one volume, one Postgres database and one bucket prefix. Postgres listens on localhost only. There is no shared pool and no per-tenant DNS record to leave behind. One KMS key per tenant encrypts the volume, database and backups. Deleting a tenant destroys the key first, wipes the volume, purges the bucket prefix and emits a signed receipt within 30 days. Backups age out under the same key, so they become unreadable immediately.

Tenant instances have no public SSH. Administrative access is SSM Session Manager, recorded in CloudTrail. Support access is a tenant-issued, time-boxed grant, visible in the tenant audit log, not a permanent account. Break-glass requires TOTP plus a one-time code to the verified owner, is scoped to one IP, expires in eight hours or less, is written to the audit log and auto-expires. The owner approves every exception.

Input plus output tokens, weighted by model. 5M billable tokens are included each month. Failed runs are not billed and cache hits are free. Overage is $15 per 1M tokens, billed monthly from the gateway’s own count, not from an estimate.

It is a product entitlement on the same workspace, and it is included in the price. A tenant can be granted tender, help desk, or both. Without the entitlement the routes return 404, so there is no half-visible menu advertising something the tenant cannot open.

Members see and reply to their own tickets. Agents see the tenant-wide queue and can reply, assign and change status. Admins manage the knowledge base: collections, imports and publishing. In this version the roles come from a deployment allowlist; a role editor is not built yet.