The Knowledge Hub

Agents are only as good as what they know

The Knowledge Hub is where an organisation's knowledge becomes usable by its agents. Everything an agent draws on when it works, product details, plans, brand assets, team files, and its own skills, lives here, organised into three levels and retrieved through a hybrid model built for both precision and speed.

Knowledge on Nagent is scoped, not pooled. An agent sees only the knowledge its level and team permit, and every retrieval respects the tenant boundary: nothing in one organisation's Knowledge Hub is ever visible to another.

A hybrid retrieval model

Different kinds of knowledge need different retrieval. The Knowledge Hub runs two systems side by side and routes each request to the right one.

Semantic retrieval for documents, data, and instructions

Text-based knowledge, documents, structured data, plans, instructions, and policies, is indexed into a vector database and retrieved through RAG (retrieval-augmented generation). When an agent works on a task, the hub retrieves the passages most relevant to that task and supplies them as context, so agents reason over the organisation's actual material rather than general knowledge.

Vector storage runs on MongoDB, with Pinecone available as the vector layer for Enterprise customers.

Direct retrieval for media

Media assets, images, video, logos, and brand files, do not go through semantic search. They are stored and served through a CDN and retrieved directly by reference. When an agent needs the correct logo lockup or a product image for a campaign, it fetches the asset itself, exact and unmodified, rather than a description of it.

This split matters in practice: semantic retrieval finds the right passage, direct retrieval guarantees the right file. Brand-critical assets always arrive byte-for-byte as uploaded.

Three levels of knowledge

Knowledge in the hub sits at one of three levels. Each level has a different owner, a different scope, and a different job.

1. Enterprise Knowledge

Link: https://nagent.ai/admin/knowledge

The organisation-wide layer, available to every team and agent in the tenant (subject to team scoping rules).

Enterprise Knowledge holds what is true of the whole organisation: documents and data, plans, product and service details, the logo and brand assets, and brand guidelines. This is where the single source of truth lives; when an agent anywhere in the organisation describes a product or applies the brand, it draws from here.

Owned and managed by organisation Admins.

Admins can create new knowledge by:

• pasting text
• pointing towards a url
• manual upload
• fetching from a tool through API/MCP connection

Media assets

Logos, product images, video, and other brand files, stored for direct retrieval via the CDN rather than semantic search. Assets used in agent output arrive exactly as uploaded, byte-for-byte. Enterprises can point the media assets to a CDN or Storage system.

Link: https://nagent.ai/admin/image-gen/library

ICP profiles

Ideal customer profiles: who the organisation sells to, segmented and described. Marketing and sales agents use ICP profiles to target content, qualify leads, and choose the right message for the right audience.

Link: https://nagent.ai/admin/icp-profiles

Brand guidelines

The rules of the brand: voice, tone, terminology, visual standards, and usage rules. Agents producing customer-facing output check their work against this section, so the guidelines uploaded here directly shape everything agents write and design.

Nagent uses Brand Lock AI to transform the guidelines into a markdown file.

Link: https://nagent.ai/admin/brand

Competition

Competitor profiles and competitive intelligence. Agents drafting positioning, handling objections, or producing comparative content draw on this section rather than improvising claims about competitors.

Link: nagent.ai/admin/competitive-intel

2. Team Specific Knowledge

The team layer, scoped to a single AI Team.

Team Specific Knowledge holds the team's shared files, its working surface, and its system of record: the material a team's humans and agents both work from day to day. What one agent in the team learns or produces lands here, where teammates, human and agent, can build on it rather than starting over.

Owned by the team; visible only within it unless deliberately shared.

3. Agent Specific Knowledge

The individual layer, scoped to a single agent.

Agent Specific Knowledge is the agent's skills: the documented, versioned procedures that define what the agent does and how. A skill is not a general capability, it is knowledge, written down, with guardrails, escalation rules, and rollback steps included. Two agents on the same team can share every document and still behave completely differently, because their skills differ.

Skills are authored and managed in the Agent Workbench; see Creating an Agent and Skills.

How the levels combine

When an agent works, its effective knowledge is the stack of all three: its own skills, its team's shared knowledge, and the enterprise layer its scope permits. Retrieval draws across the stack, with the narrower levels providing specificity and the enterprise level providing ground truth. Conflicts resolve towards the more specific level: a team's documented way of working takes precedence over a generic enterprise document, and a skill's explicit instructions take precedence over both.

Populating the Knowledge Hub

1. Open the Knowledge Hub and select the level to add to: Public, Enterprise, a specific team, or a specific agent.
2. Add sources: upload files, connect document stores, or point the hub at sites to capture.
3. Text sources are indexed automatically for semantic retrieval; media is registered for direct retrieval. Indexing status is visible per source.
4. Review scope: confirm which teams and agents can see each source before agents begin drawing on it.

What to do next

With the Knowledge Hub populated, agents have something to know. The next step is giving them something to do: provision agents in the Agent Workbench, or install a prebuilt team as described in Setting Up the Team Space.