AI Prompt Library

Overview

The AI Prompt Library is a feature in DiligenceVault that lets your team define and store custom instructions for AI-powered tasks, starting with AI Autofill. Instead of each analyst writing ad-hoc prompts and getting inconsistent results, the Prompt Library makes your firm's drafting standards explicit and applies them automatically, every time Autofill runs.

When an analyst runs AI Autofill on a question or template, the relevant saved prompts from the library are applied on top of DiligenceVault's standard autofill logic. The result is draft responses that are more consistent, more predictable, and aligned with how your firm wants the work done, regardless of who runs it.

Prompts in this release are accessible to your entire firm. Per-user prompts are planned for a future release.

Who this is for:

Requestors using AI Autofill to generate first drafts of IC memos, ODD summaries, manager profiles, document extractions, etc
Responders using AI Autofill to draft responses to standard DDQs in their own DiligenceVault space

How Prompt Guidance Works

Prompts are set at three levels. Each level narrows the scope of the guidance. You can use them together to build layered, precise instructions.

Project Level : Use this for firm-wide style rules that should apply across the board. It can set the default tone, structure, and format for every response in a project.

Example: "Respond in third person, present tense. Refer to the firm as 'the Firm', not 'we' or 'our'. Use US English spelling, no contractions. Spell out acronyms on first use."
Section Level : Applies subject-matter guidance to a specific category or subcategory, such as Operations, Compliance, or Investment Process. Use this when a particular topic area requires specific sourcing rules or standards.

Example: "For Operations questions about BCP, disaster recovery, and cyber events, draw from operational docs and audit/attestation reports.
Response Level : Targets a single question with precise rules. Use this when accuracy depends on which specific data to pull, how to resolve conflicts between sources, or what level of detail to include.

Example: "Return the firm's full legal name verbatim, including entity suffix (LLC, LP, Inc., Ltd.). Look in regulatory filings or firm overview docs first. Do not abbreviate or expand."

Project guidance sets the baseline. Section guidance refines it for the relevant topic area. Response-level guidance overrides where the question requires exact handling. The rule closest to the question wins.

What You Can Guide

Prompts can shape a wide range of output characteristics. Common categories include:

Configuration

Autofill prompts are configured in the Template Builder.

To add a prompt, navigate to Diligence > Template > Select a template > and click on set automation

Selecting Set Automation in the template toolbar opens a small dropdown with two options:

Set Template Level Prompt: Opens the Create Prompt panel where you can create and save a new prompt at the template level.
View All Template Prompts: Displays all prompts already saved against the current template, so you can review, edit, or delete existing prompts.

Creating a Prompt

When you select Set Template Level Prompt, the Create Prompt panel opens. Complete the following fields to define your prompt:

Type: Select the prompt type from the Type dropdown. Auto-fill is the current supported type and is selected by default.
Name: Enter a descriptive name that identifies the purpose of the prompt, such as 'IC Memo Tone' or 'AUM Extraction Rule'. This name appears in the Applied Prompts section of response history.
Prompt: Enter your custom instructions in the Prompt field. This is the core of your prompt — define the specific guidance for the AI, such as tone, sourcing rules, format, or terminology requirements.
Add from Library: Selecting Add from Library lets you copy a previously created prompt directly into the Prompt field. This is useful when you want to reuse instructions that have already been authored and saved in your firm's library, without rewriting them from scratch.
Start with a template: Selecting Start with a template loads a sample prompt into the Prompt field with placeholder guidance covering common categories such as tone, format, emphasis, and word count. You can edit this sample to match your firm's requirements.
Advanced options : Expanding the Advanced options section reveals an Add as universal prompt checkbox. Selecting this checkbox saves the prompt as a Universal Prompt in your firm's library and links it to the current template at the same time.

Universal Prompts

A Universal Prompt is a single shared prompt definition that can be linked to multiple templates, projects, sections, or questions. It acts as your firm's single source of truth for a given standard.

Key behaviors:

One standard: Define firm-wide drafting rules once, centrally. Link the same definition to as many templates and questions as you need.
Update once, everywhere: Update a Universal Prompt and every linked template, project, and question inherits the change automatically. No manual sync, no version drift.
Reuse without recreating: Link a single Universal Prompt across multiple templates, categories, sub-categories, and questions. Different contexts, same standard.

How Conflicts Resolve

When multiple prompts apply to the same response, the following rules determine which takes precedence:

Specificity wins: When prompts at different levels disagree, the most specific level wins. Response-level beats section. Section beats project. The rule closest to the question decides.
Recency wins: If a platform instruction and a custom prompt conflict, the most recent one applies.
No conflict: If a platform instruction and a custom prompt do not conflict, both are applied.

Built-In Attribution

The AI reasoning mentions the impact your custom prompts had and the response history records which prompts contributed to a drafted answer. Expand the reasoning panel and ‘Applied Prompts’ section in the history to see the exact prompt texts applied and from which levels. Reviewers can verify what guided the AI.

Scope

Example Use Cases

For Requestors

IC Memo First Draft

An allocator team running ODD on a new manager can use AI Autofill to generate a first-draft IC memo from manager facts, DDQ responses, scores, flags, and uploaded documents. At the project level, they can set a prompt such as: “Write in third person, present tense, using a professional investment committee tone. Avoid contractions and promotional language. Clearly distinguish facts, diligence observations, and open risks.” At the section level, the Investment Process section can use: “Summarize the manager’s investment philosophy, sourcing, due diligence, portfolio construction, risk management, and sell discipline. Draw from the PPM, DDQ, and Form ADV only.” This helps the memo draft follow the firm’s IC format without the analyst having to manually restructure every section.

Executive ODD Summary

An ODD team can use the Prompt Library to create a consistent 5-page executive summary for every manager review. A Universal Prompt can be used across all ODD summary templates, such as: “Create an executive-ready ODD summary with the following sections: Overview, Key Strengths, Key Risks, Mitigants, Open Items, and Recommended Follow-Ups. Keep the language concise, factual, and risk-focused.” For the Risk section, a section-level prompt can say: “For each risk, explain why it matters, identify any mitigating controls, and flag where evidence is missing.” This ensures summaries are consistent across managers and analysts.

Document Extraction from AFS, PPMs, ADVs

Teams can use prompts to tell AI Autofill exactly which documents to use for different types of information. A project-level prompt can say: “Extract structured data only from the documents available in the project and include source citations wherever possible. Do not infer missing information.” A section-level prompt for fund terms can say: “Use the latest PPM or LPA to extract management fee, performance fee, hurdle, liquidity, lock-up, redemption notice period, gates, and fund expenses.” A response-level prompt for AUM can say: “Use the most recent audited financial statement first; if unavailable, use Form ADV. Include currency and as-of date.”

Manager Profiles Pre-Filled by Firm Rules

Allocator and consultant teams can use prompts to pre-fill manager profiles before sending them to managers for review. A project-level prompt can say: “Populate profile fields using the most authoritative source available. Use consistent formatting for firm name, AUM, dates, addresses, service providers, and key personnel.” A response-level prompt for legal name can say: “Return the full legal name exactly as stated in the most recent regulatory filing, including entity suffix. Do not abbreviate.” For AUM, the prompt can say: “Use the latest AUM figure from the AFS, Form ADV, or firm overview, in that order. Include currency and as-of date.”

For Responders and Asset Managers

Standard DDQ Responses for Asset Managers

Asset managers can use Prompt Library to draft DDQ responses in their approved voice and structure. A project-level prompt can say: “Respond on behalf of the firm using formal, client-ready language. Use approved firm terminology and avoid unsupported claims, guarantees, or promotional statements.” A section-level prompt for Investment Process can say: “Describe the firm’s investment philosophy, research process, portfolio construction, monitoring, and risk management using approved materials where available.” A response-level prompt for AUM can say: “Use the latest approved firmwide AUM figure. Include as-of date and currency. Do not use strategy-level AUM unless specifically requested.”

Saved Prompts for Tone, Structure, and Reuse

Teams can save commonly used instructions and reuse them across templates, projects, sections, or questions. A reusable prompt could be: “Start with a direct answer, then provide supporting detail. Use clear paragraphs unless the question asks for a list. Keep the tone professional, factual, and concise.” Another saved prompt could be: “Avoid marketing language, unsupported superlatives, and assumptions. If source material does not contain the answer, state that the information was not found.” This helps every analyst get consistent output without rewriting the same instructions each time.

Tips

Start with a project-level prompt to set baseline style, then layer section and response-level prompts for areas that need more precision.
Use Universal Prompts for firm-wide standards that should never drift, such as entity name formatting or preferred terminology.
Check the Applied Prompts section in response history after running Autofill to verify your prompts are being applied as expected.
If a response does not look right, review whether a more specific prompt at the response level is needed to override a broader instruction.