Commercial pilot v0.1.0

RFP Analyzer

Pre-processing of incoming RFPs (public B2G + private B2B + framework agreements): RFP extraction, formal eligibility, gap matrix with verifiable citations of catalog and past references, draft response by sections, bid / bid_partner / no_bid recommendation. The agent never submits — the bid manager finalizes.

See the live demo Source code on GitHub

Measured gain

1-1.5 ETP

veille + pré-mapping AO récupérés / +5-10 pts taux de gain

For a mid-market firm handling 200 RFPs/year: 1 to 1.5 FTE of watch + pre-mapping recovered (~€120k/year), plus +5 to +10 points of win rate because ineligible RFPs no longer distract the team. The partner consultancy that sells it uses it on itself first — shortest sales cycle in the catalog.

The problem

RFP response in mid-market is a mechanical-work faucet that costs revenue in parallel with the revenue it generates.

Today

Pré-analyse manuelle

Un commercial parcourt BOAMP/TED et lit 150 pages de CDC

7-20 j-h / AO (analyse + brouillon)

60-70 % d'AO non-éligibles ou non-pertinents

→ 1-2 ETP de veille commerciale en pure perte

· Lecture de 60-150 pages de cahier des charges
· Coche manuel des exigences contre le catalogue
· Recherche manuelle des références passées les plus proches
· ChatGPT Enterprise = impossible (prix négociés, marges sensibles)

Les AO non-éligibles distraient des AO gagnables

With RFP Analyzer

Pré-analyse assistée

Sur l'appliance, le commercial finalise et dépose

Quelques min. / AO (extraction + éligibilité)

0 référence inventée

→ 1-1.5 ETP veille + pré-analyse récupérés (~120 k€/an)

› Éligibilité formelle auto (CA N-1, ISO, HDS, SecNumCloud)
› Matrice de couverture avec citation vérifiable (Layer D grounding)
› RAG ne trouve pas → section « à compléter manuellement »
› L'agent recommande, le commercial / responsable d'offre soumet

+5 à +10 points de taux de gain

How it works

Four steps, from incoming document to human decision.

1

Structured RFP extraction

`extract_rfp_sections` splits the PDF into sections: administrative requirements, technical requirements per lot, weighted evaluation criteria, planning, pricing terms. Without this output, the agent goes to `hold_for_review` — no attempt to extract requirements 'as best it can' from the raw text, an RFP is a legal document.
2

Formal eligibility

For each administrative requirement (revenue, ISO/HDS/SecNumCloud certs, professional liability insurance, minimum past references, Defense clearance), `check_eligibility_criteria` compares against the firm's actual state. Any eliminating criterion unmet → `eligibility_check.status: blocked` → forced `no_bid` recommendation. Confidential-Defense / OIV RFPs without clearance are killed in 30 seconds rather than going to a meeting.
3

Gap matrix with verifiable citations

For each technical requirement, `lookup_catalog` searches the service catalog and `lookup_past_references` searches the won-RFP archive. Every `requirement_id`, `catalog_entry.id`, `past_reference.id` in the output is cryptographically bound to a tool call this turn (Layer D strict grounding). The agent **cannot** cite a reference that doesn't exist — grounding rejects the output.
4

Indicative pricing + draft response + bid/no_bid recommendation

`pricing_skeleton` builds line items from man-days × standard TJM pulled from the catalog. No margin calculated, no suggested sale price — the commercial / CFO arbitrates. `response_draft.sections[]` produces markdown per section (executive summary, methodology, team with GDPR-anonymized CVs, planning, cited references, indicative pricing, compliance). `bid_no_bid` recommends bid / bid_partner / no_bid / hold_for_review with confidence score and objective list of risks. The COMEX decides.

Architecture

Tools, connectors, deployment.

Tools (function-calling)

extract_rfp_sections
lookup_catalog
lookup_past_references
lookup_team_profiles
check_eligibility_criteria
escalate_human

Optional connectors

commercial-rfp-parser
commercial-catalog
commercial-references
commercial-cv-database
commercial-eligibility
commercial-bid-queue

Each connector activates based on the customer's subscription.

Deployment via the SDK

$ lmbox agent deploy ./rfp-analyzer \
    --box BOX-XXX \
    --token "$LMBOX_BOX_API_KEY" \
    --api https://api.lmbox.eu

LMbox guarantees across the catalogue

Data stays with you

Model and data stay on the customer's LMbox appliance. No patient, contract or invoice data is ever sent to an external cloud.

Audit chain

Every tool call, every agent output is timestamped, hashed and admissible before the regulator (ACPR, ANSM, CNIL, EBA).

Human decision

The agent recommends, the human decides. No auto-signature, no auto-payment: final responsibility stays with the business.

Other catalogue agents

Law firms pilot

NDA Reviewer

Reads each incoming NDA, identifies non-standard clauses against the firm's internal template library, drafts an amendment memo for the responsible partner.

1 h 30

Measured gain · par contrat

2 tools 2 connectors

v0.1.0 See the agent

All sectors pilot

Meeting Summarizer

Turns a raw meeting transcript (Teams, Zoom, locally-transcribed audio) into a structured CR: decisions made, action items with owner and deadline, open questions.

20-40 min

Measured gain · par réunion

3 connectors

v0.1.0 See the agent