v1.0.0-rc3, AI agents can query Knowledge Nexus spaces directly from the prompt, using the native knowledge_nexus_search tool. This allows answering based on internal documentation, FAQs, and knowledge bases maintained in Nexus, without configuring a custom tool.
For the CRM’s built-in knowledge base (PDFs, manually pasted files), keep using the Knowledge Base module. Nexus is complementary — useful when your organization already maintains a separate knowledge base and wants to reuse it in the agent.
Prerequisites
- An Evo Nexus installation reachable from the CRM network.
- At least one space created in Nexus with indexed documents.
- Permission in your role to edit agents (
agents.manage).
Linking a Space to the Agent
- Open an agent’s detail at
Agents → [agent name]. - Go to Integrations → Knowledge Nexus.
- Click the space picker. The CRM queries the Nexus API (via the core-service proxy) and lists available spaces.
- Check the spaces this agent should be able to query.
- Save.
How the Agent Uses Nexus
The agent automatically receives a tool calledknowledge_nexus_search in its vocabulary. When the user asks a question whose answer is plausibly in the base, the LLM calls this tool:
Expected Behavior
- Answer from source: if the found snippet answers directly, the agent quotes the information.
- Answer with gap: if the base does not cover the case, the agent reports it did not find specific information and asks for more context, instead of making things up (with a well-calibrated prompt).
- Empty return: if the search returns nothing, the agent continues the conversation without citing the base.
When to Use Nexus vs CRM Knowledge Base
It is not exclusive — you can use both in the same agent.
Diagnosing
If the agent is not using Nexus despite being linked:- Confirm the link. Reopen the agent detail; the space should appear checked.
- Check permissions. The agent must have
agents.readon the linked resources. - See if the space has indexed documents. If the space is empty or still processing documents, the tool returns empty.
- Review the prompt. If the agent’s system prompt is too rigid (e.g., “answer only based on what you know”), it may not call the tool. Adjust the prompt to make explicit when to consult the external base.
- Processor logs. The
evo-ai-processor-communityservice logs tool calls — useful when you need to confirm whether the tool was actually invoked.