Ground agents in approved company knowledge—without turning the whole workspace into context.
Attach scoped knowledge bases to configured agents, control when retrieval can activate, and keep company knowledge distinct from agent memory and operational systems of record.
A knowledge base is a governed retrieval resource—not a bigger system prompt.
A production agent should not receive every document a workspace owns on every run. Knowledge retrieval selects relevant information from configured sources when the current agent, context and conditions allow it, then provides that material to the model as working context.
Company material the agent may retrieve when the resource is in scope.
Bring relevant source material into context instead of loading an entire corpus.
Knowledge bases can remain scoped and conditionally available rather than globally inherited.
Knowledge, memory and systems of record solve different problems.
Treating all context as one bucket creates stale facts, accidental leakage and unclear authority. Createting keeps these concepts separate so each can have a different trust and lifecycle model.
Policies, manuals, product information, procedures and other approved source material.
Persistent context whose read and write behavior needs its own scope and policy.
Explore agent memory →CRM, ERP, ticketing and transactional systems should remain authoritative for business records and actions.
Knowledge should be assigned, not silently inherited.
Agent Studio can attach knowledge bases to reusable agent configurations. Conditions and runtime context then determine whether a given knowledge resource is eligible for a run. That keeps one support agent from automatically receiving the same corpus as a finance, sales or research agent.
The resource belongs to the configured organizational context.
Only agents configured with the resource should treat it as available knowledge.
Event, data, context, time, system, approval or reasoning conditions can narrow when the resource activates.
Only relevant material should enter the model context for the current work.
RAG is a retrieval path, not a trust decision.
Retrieval-augmented generation helps a model answer from external information instead of relying only on model parameters. But retrieval quality alone does not decide whether the source should be visible to this agent or whether a retrieved statement is authoritative for a downstream action.
Exact retrieval strategies and thresholds depend on the configured deployment and knowledge backend.
Knowledge can inform an action. It should not authorize the action.
A retrieved policy may tell an agent what a refund rule says. Tool governance still decides whether the agent can actually issue a refund, whether a human must approve it, or whether the action is denied.
One configured knowledge resource can support multiple execution surfaces.
When the same agent configuration is used across chat, voice, missions or other enabled runtime surfaces, the knowledge relationship stays attached to the agent instead of being rebuilt separately for each channel.
Use retrieval for changing knowledge. Use model adaptation for a different problem.
Knowledge bases can update the information available to an agent without changing the model weights. Fine-tuning or private model adaptation is a separate infrastructure decision for behavior, specialization or deployment requirements—not the default way to keep company facts current.
Good fit for policies, product documentation, procedures and other evolving source material.
A separately scoped model and infrastructure problem.
Explore AI Infrastructure →AI agent knowledge bases
What is an AI agent knowledge base?
It is a configured information resource that an AI agent can retrieve from at runtime so relevant company knowledge can enter the model context without placing the full corpus in every prompt.
How is a knowledge base different from agent memory?
Knowledge bases usually contain curated or connected organizational information. Agent memory is persistent reusable context created or updated through prior interaction or execution and therefore needs separate read and write policy.
Does Createting use RAG?
Createting supports knowledge retrieval as a configured agent resource. The specific retrieval backend and strategy depend on the deployed knowledge configuration.
Can every agent access every knowledge base?
No. Knowledge bases are intended to remain scoped resources attached to the relevant agent configuration rather than universal workspace context.
Can knowledge access depend on conditions?
Yes. Knowledge bases can use event, data, context, time, system, approval or reasoning conditions before activation.
Does a knowledge base replace CRM or ERP data?
No. Knowledge retrieval can inform reasoning, while transactional systems should remain authoritative for operational records and state.
Map the knowledge an agent needs—and the information it should never receive by default.
Bring one agent workflow, its data sources and its trust boundaries. Createting can help separate knowledge retrieval, memory, tools and systems of record.