The human-approval layer

Strategy Lock.
The full walkthrough.

Most AI bid tools run end-to-end automation. BidEngine inserts a human approval step between intelligence and drafting — every win position, ranked priority, must-prove claim, tone and avoid-list approved by the bid team. Signed, dated, version-stamped, and propagated downstream.

The “guardian agent” pattern Gartner is calling — built natively into the bid workflow, not bolted on after the fact.

The flow

Platform proposes

Intel + Gate

Buyer profile, win position, deal-breakers — auto-drafted from the tender pack.

Bid team owns

Strategy Lock

Review, refine, lock. Versioned audit row. Live evidence readiness checked.

Platform executes

Write + Refine + Check

Drafts, scores and audits — all against the approved strategy. Every output stamped with the version.

The four mechanisms

01

Platform-prepared, team-locked

An approval screen between BidGate and BidWrite. Auto-populated with win position, ranked priorities, differentiators, risks. Drag-rank, edit, lock. Refine — don't write from scratch.

02 · Signature feature

Live evidence readiness

Every must-prove claim runs a live BidVault search the moment you type it. Three states — see the evidence gap before you lock, not three days from submission.

✓ strong⚠ weak✗ gap

03

Versioned audit trail

Every lock event creates a permanent, immutable row. Re-locking creates v2.0 — prior versions preserved, status-flagged superseded. Full chain visible: who locked what, when, with which position.

v1.0v2.0v3.0 · live

04

Version label propagates

Every BidWrite draft, BidCheck score and audit chip carries the strategy version that produced it. The model prompt header reads “APPROVED STRATEGY v2.0 — AUTHORITATIVE.” Trail visible to user, model, and regulator.

BidEngine is built around a versioned human-approved strategy layer. Every drafted answer carries the strategy version that produced it.

Private Cloud (VPC) · Enterprise deployment

When tender data can't leave your network.

Many public-sector, defence and regulated buyers prohibit confidential bid material from touching third-party AI services — which is exactly why pasting tenders into general tools like ChatGPT or Gemini gets banned by security teams. For those organisations, BidEngine can be deployed single-tenant, inside your own Virtual Private Cloud (VPC).

The same eight-specialist platform and the Strategy Lock layer above — but the documents, the evidence and the AI inference all run within your boundary. Nothing is sent to an external AI provider, and no tender content leaves your network.

What stays inside your boundary

In-cloud AI inference

Claude runs via your own cloud (AWS Bedrock / private endpoint) — no calls to public AI APIs.

Your storage, your keys

Uploaded documents and the evidence vault live in your buckets, encrypted, under your control.

No model training

Your bid data is never used to train any AI model — contractually and architecturally.

Single-tenant isolation

A dedicated instance for your organisation alone — no shared infrastructure, no co-tenants.

Your identity provider

SSO / SAML sign-in into your existing identity stack.

Your network perimeter

Deployed in your VPC behind your firewalls and egress controls — auditable end-to-end.

Private Cloud is a bespoke, contracted deployment — scoped to your environment and security requirements.

Talk to us about Private Cloud