OpenBid turns a consultancy’s 14 years of accumulated knowledge into a living, governed, agentic capability — so every proposal team can find what the firm already knows, identify the right people, and produce a credible first draft in hours, not days.
Grounded · cited · access-controlled · human-gated. Trust is the binding constraint — and we engineered for it. The model never receives a document the user isn’t entitled to see.
We have delivered comparable financial-services outcomes and can reuse accredited assets, shortening time-to-value… [d0001 · proposal]
Competitor (Sterling Union) material was filtered out of retrieval before the model saw it. Every cited asset is one this bid team is entitled to reuse.
14 years of deliverables sit in SharePoint and local drives. Expertise data goes stale in a CRM. There is no shared research repository. When a bid arrives, proposal teams spend 60–70% of their time searching — and credible-but-lost opportunities cost millions a year. The core thing AI must solve is not “write proposals” — it is making the firm’s collective memory instantly and safely usable at the moment of a bid.
Grounded, cited, access-controlled by construction — and measured every release.
A supervisor agent decomposes a bid into a plan, then delegates to Knowledge Retrieval, Expertise Routing, Proposal Drafting, Market/Freshness and Governance agents — parallel reads, sequential writes.
Dense + sparse fusion over 14 years of proposals, deliverables and methodologies. Every passage returns with its citation, a confidence score, and a stale flag.
Who-knows-what is mined from the work itself, not a stale CRM. Ask 'who's done this, in this sector, at this seniority' and get ranked people with evidence.
Access control is a pre-retrieval query filter — the model never receives a chunk the user can't see. Confidentiality tiers, Chinese walls, PII scrub, full audit.
Time-decay by document type — market intel decays in months, methodology in years. Stale content surfaces with a badge and a ranking penalty, never silently.
OpenAI Codex SDK, Claude Agent SDK, or a dedicated RAG agent with hard guardrails — all sharing one RAG-over-MCP tool surface. Autonomy where safe, structure where it matters.
Plus a deterministic guardrail agent for the safety-boundary cases.
Chunk + embed + retrieve. Useful for Q&A; can't route people, draft, or govern access.
Hard-coded DAGs. Brittle. Hard to evolve as the knowledge base and policies change.
Claude-Code-style — read the corpus, plan, call tools — paired with a deterministic guardrail agent for the cases where structure must beat autonomy.
Connect SharePoint / drives. Documents are parsed, typed, chunked, embedded and stored with governance + freshness metadata in pgvector.
Authorship and staffing signals derive a self-maintaining graph of who-knows-what at concept level — no admin burden, no stale CRM.
A bid brief fans out to retrieval, expertise and market scans in parallel; drafting, governance and review run as one auditable chain.
A cited draft with a reuse rate and confidence — human-gated at plan, shortlist and review. Below confidence or across a wall, it escalates, never fabricates.
Our test corpus has three planted traps. The platform catches all three — verified, re-runnable, in every release scorecard.
A 2019 market outlook is surfaced with a stale badge — never reused as if it were 2026 truth.
A restricted M&A document is invisible to everyone — including the Managing Partner — except the entitled deal team.
A competitor's proposal can never cross the Chinese wall to the rival client's bid team, even when explicitly requested.
Ashford Partners is the reference tenant — strategy, operations and technology advisory across four sectors.
Open banking, core-banking migration, payments modernisation, regulatory remediation — repurpose prior bids and surface the right regulatory specialists.
Net-zero transition, smart metering, grid data platforms — with confidential M&A work walled off to the deal team by need-to-know.
EHR implementation, patient-flow optimisation, workforce planning — reuse accredited methodologies and clinical case studies.
Digital identity, procurement transformation, shared-services operating models — credible first drafts grounded in real prior delivery.
Month 1: a baseline and adoption. Month 6: search time and draft time falling, reuse climbing. Year 2: a win-rate and bid-cost number — measured against a control cohort, with caveats you can audit. Framed as capacity and win-rate, not “AI”: every proposal hour saved is a billable hour returned; a point of win-rate on a £3M bid is ≈ £30k.
AGPL-3.0 community license. Commercial license for firms with revenue ≥ £250k or air-gapped deployments.
Open the live demo — pick a persona, ask a question, and watch the platform ground its answer in entitled material, flag what’s stale, and refuse to cross a confidentiality wall.