Knowledge Management for Climate Finance

CFI operates on a zero-retention architecture — no prompt or session data is retained or logged after a query completes. All data is processed strictly in-memory to generate a response and then permanently discarded. No human review occurs, and no data is used for model training. The platform is hosted on dedicated VPS infrastructure in a certified German data centre with AES-256 encryption at rest and TLS 1.2+ encryption in transit. Each client operates on a dedicated instance — no shared infrastructure with other clients. Google has granted Janus a formal, written exception to its standard prompt logging policy, ensuring zero retention at the AI inference layer as well. Self-hosting on your own cloud infrastructure is also available.

CFI is delivered as a professional services engagement, not a software subscription. Enterprise pricing is estimated at $25,000 USD per year, structured across a four-year engagement period — a total programme cost of $100,000 USD. The client retains full ownership of all configured tools, knowledge layers, and documentation throughout and beyond the engagement. On-demand analytical services — such as individual Concept Note evaluations and co-financing benchmarks — are available separately and independently of the enterprise engagement.

Yes. CFI is positioned as an institutional capacity-strengthening activity under GCF Readiness and Preparatory Support funding. GCF's own Readiness Guidebooks explicitly fund digital Knowledge Management Systems for Direct Access Entities, and CFI is purpose-built to meet those requirements. Several Accredited Entities are currently incorporating CFI into their readiness proposals.

CFI integrates three knowledge layers:

• Layer 1 - Donor Knowledge: The complete approved GCF corpus, indexing over 3,000 project files including Concept Notes, Funding Proposals, PPF applications, Board decisions, and framework papers.
• Layer 2 - Climate Science: Approximately 100+ verified open-access publications from the most authoritative bodies in the field (IPCC, World Bank, AfDB, UNEP, WMO, WHO) — the sources most frequently cited in approved GCF submissions.
• Layer 3 - Institutional Knowledge: Your organization's own submission history, internal technical notes, and project files — preventing the loss of institutional memory when staff rotate or consultants depart.

One query searches all three layers simultaneously.

The platform's donor knowledge layer is built on the complete GCF corpus, and coverage extends to the Adaptation Fund and GEF. AI assistants are calibrated to GCF and Adaptation Fund submission standards. Additional donor knowledge layers — such as AfDB, EIB, World Bank, and bilateral climate funds — are available on a separately scoped basis.

No. CFI does not generate AI content on behalf of clients for submission to the Secretariat. It provides evidence-based intelligence and quality assurance to support your own preparation and review processes. The platform empowers your team with the knowledge and benchmarks to produce stronger submissions — the drafting and decision-making remain with your team.

CFI includes task-specific agentic AI assistants configured and delivered as part of the professional services engagement:

• Primary Insights Search: Surfaces comparable approved projects with their theory of change, co-financing structure, and design features.
• First Draft CN Builder: Produces a structured first-draft Concept Note aligned to GCF or AF templates, anchored in approved precedent.
• Concept Note Reviewer: Stress-tests a draft CN against comparable approved CNs, flagging gaps in climate rationale, theory of change, and investment criteria.
• Funding Proposal Reviewer: Section-by-section audit of any FP against the full corpus, ranking corrective actions by scoring impact.
• GCF Reviewer Response Assistant: When the Secretariat sends review comments, surfaces how comparable approved proposals addressed the same issues.

The platform also includes several hundred pre-built prompts for Environmental and Social Safeguards (ESS) and Monitoring, Evaluation, Accountability and Learning (MEAL) workflows.

CFI uses a Retrieval-Augmented Generation (RAG) architecture that grounds every response in curated, verified source documents. Responses cite specific approved projects and publications from the knowledge layers — not generated text. AI assistants are configured with constrained output structures and evaluation criteria drawn directly from donor guidance. This architecture significantly reduces the hallucination risk associated with general-purpose AI tools. The human-in-the-loop approach ensures that domain experts review and act on insights, not raw AI output.