Skip to content
Search icon
Book a demo

Agent Intelligence Service

A managed intelligence layer that gives AI agents immediate, curated direction at the point of task execution. Replacing guesswork, web searches, and unverified outputs with maintained, persona-specific knowledge that's ready to use the moment an agent needs it.

It works with any AI tool or large language model. Whether you're building agents for enterprise workflows, deploying specialist AI assistants, or extending an existing platform, Agent Intelligence Service provides the layer of current, verified direction that general-purpose AI cannot reliably supply on its own.

The Problem It Solves

Modern AI tools are highly capable at processing large volumes of information such as reading documents, extracting data, identifying patterns, and generating structured outputs. But many professional tasks require more than processing power. They require direction: a clear, current understanding of what a specific audience, persona, or stakeholder is actually looking for.

 

Without that direction, an agent has two options: rely on training data that may be months or years out of date, or spend time searching the web and attempting to verify what it finds before it can usefully apply that knowledge. Neither is reliable. Neither is fast. And for business-critical tasks, neither is good enough.

Lens-Grounding

Agent Intelligence Service introduces lens-grounding - providing an agent with a curated, up-to-date intelligence lens that immediately answers the question: "what does this type of person care about, right now?"

Rather than asking an agent to research a stakeholder's priorities, verify its findings, and then apply them to the task at hand, Dovetail has done that work upfront. At the moment an agent runs a task, it requests the relevant lens from the Agent Intelligence Service and receives immediate, verified direction.

Lens-Grounding vs. Knowledge Grounding

These are complementary capabilities that solve different problems and understanding the distinction matters. 

 
 

Knowledge grounding

 

Connects an agent to specific information sources: your internal documents, your databases, your organisational knowledge base. It solves the problem of access, ensuring the agent answers from your data rather than from general training. It is highly effective and widely used, and it is not what Agent Intelligence Service does.

Knowledge grounding says: here is the information to answer from.

 

Lens-grounding

 

Solves a different problem entirely. It is not about what information the agent can access, it is about how the agent is directed to approach the task. The agent may already have everything it needs: the contract is there, the financial plan is there, the project report is there. What it lacks is clear, current, expert-level direction: what does a PE investor look for in this document right now? What risk patterns is a procurement specialist trained to spot? What would an auditor flag first?

Lens-grounding says: here is how to approach the question and what to prioritise.

Used together, they represent a complete grounding strategy. Used independently, lens-grounding addresses the quality and currency of the agent's analytical direction, the dimension that generic AI most consistently underserves.

How It Works

1

The agent identifies the task type and the relevant lens - for example, Contract Stress Test + In-house counsel perspective, or Financial Plan Review + CFO lens, or Supplier Evaluation + Procurement Risk lens

2

It sends only minimal metadata to the Agent Intelligence Service - the confidential source data never leaves the client's system

3

The service returns a curated intelligence payload: the current priorities, concerns, red flags, and focus areas for that specific lens

4

The agent uses this to direct its analysis, producing output grounded in current, persona-specific intelligence rather than generic AI reasoning

This is the key architectural principle: Agent Intelligence Service adds intelligence to an LLM query without needing to see the underlying data. The confidential content - the contract, the financial plan, the project report - stays entirely within the client's own environment. Only the question of what to look for travels to the service. The answer comes back as direction. The analysis happens locally.

Maintained Intelligence, Not Static Training Data

The lenses in Agent Intelligence Service are actively maintained by Dovetail. When stakeholder priorities shift, when regulatory environments change, when economic conditions alter what an auditor focuses on, the service is updated. Agents using the service always draw on current intelligence, not a snapshot embedded in a model's training weights from months or years ago.

Example Lenses

  • Investor → Angel, Venture Capital, Private Equity
  • CFO → cash flow resilience, covenant compliance, working capital management
  • Auditor → going concern, material misstatement, estimation uncertainty
  • Customer / Buyer → liability exposure, exit rights, SLA enforceability
  • Supplier / Vendor → payment risk, scope creep, IP protection
  • Employer → restrictive covenants, compliance obligations, termination grounds
  • Employee → notice and severance, post-termination restrictions, clarity of obligations
  • M&A Due Diligence → change of control triggers, assignment provisions, undisclosed liabilities
  • Procurement Manager → total cost of ownership, supplier resilience, compliance requirements
  • Finance Controller → budget impact, payment terms, cash flow implications
  • Risk Manager → supply chain concentration, ESG obligations, single-source exposure
  • Category Manager → market benchmarks, negotiation leverage, alternative supplier landscape
  • Sub-types by category: SaaS & software, Professional services, Goods & manufacturing, Facilities & estates
  • Project Sponsor → strategic alignment, ROI visibility, milestone confidence
  • PMO / Programme Office → governance gaps, resource conflicts, dependency management
  • Client / Stakeholder → delivery confidence, change management, value realisation
  • Assurance / Audit → controls adequacy, documentation standards, compliance with methodology
  • Sub-types by delivery approach: Agile, Waterfall, Infrastructure & capital, Digital transformation

The Privacy Principle

Agent Intelligence Service is built on a simple but important architectural decision: no confidential data needs to leave the client’s system.

The service does not read your contracts, financial plans, procurement documents, or project reports. It receives only the minimum metadata needed to identify which intelligence lens to return: a task type and a lens identifier. The agent performs all substantive analysis locally, using the intelligence payload purely as direction.

This means organisations can benefit from continuously maintained, expert-level intelligence without any trade-off on data security or sovereignty. The analysis stays inside. The direction comes in.

 
 

Availability

Built into specialist Dovetail agents.

Specialist agents in the Dovetail portfolio are powered by Agent Intelligence Service, it’s what makes them demonstrably more capable than a general-purpose AI prompt for the same task. The difference is not just the output format; it is the quality and currency of the direction the agent is working from.

 

Available as an API and MCP server for third-party agent builders.

Organisations and developers building their own AI agents can integrate Agent Intelligence Service directly, drawing on Dovetail's maintained lens library or working with Dovetail to build custom lenses for their specific domain, industry, or workflow.

 

Why Lens-Grounding Changes the Quality of the Answer

Without Agent Intelligence Service

Agent reasons from general training data

Stakeholder priorities inferred, not known

Web search required to fill knowledge gaps

Verification of retrieved information adds latency

Confidential data may need to leave the client's system for enrichment

Same output regardless of who is asking

With Agent Intelligence Service

Agent is directed by current, maintained intelligence

Stakeholder priorities explicitly provided for the current period

No search required – direction is pre-curated and immediate

Intelligence is pre-verified by Dovetail

Confidential data stays local – only task type travels

Output shaped by the specific lens and sub-type