Every unanswered question becomes a reviewable improvement.
AutoLearn monitors conversations, finds knowledge gaps, and suggests grouped answers for you to review. Nothing enters the base without approval.
Enable AutoLearnWhat changes in the operation
No agent launches covering every detail of the operation. The challenge is quickly discovering which real questions did not get a good answer. AutoLearn watches conversations, identifies gaps, groups similar questions, and proposes base content for review. Your team decides what goes in, edits when needed, and turns recurring misses into reusable knowledge.
Detection from real usage
AutoLearn monitors the agent in operation and identifies when a question points to a base gap or insufficient answer.
Grouping by similarity
Semantically close questions become one item with a count, so your queue shows priority instead of noise.
Editable suggestion before approval
For each gap, AutoLearn suggests a starting answer. You approve, edit, or dismiss it before adding it to the base.
Human control in the learning loop
The feature works on agents with the “Consult Knowledge Base” tool active, but the final content decision stays with your team.
How it works
Gaps appear, get grouped, and become approved content for the base.
The gap appears in a conversation
Example: several people ask how to export to Excel, but the base does not yet have a clear answer.
AutoLearn groups and suggests
Similar questions go into the same item, with an occurrence count and a suggested answer for review.
You approve what becomes knowledge
Edit the suggestion if needed and add it to the base. In the next conversation, the agent has a source to search.
Frequently asked questions
-
Does AutoLearn work with any agent?
It works with all agents (internal and external), but requires that the "Consult Knowledge Base" tool is active on the agent and at least one knowledge base is selected. Without an active base, AutoLearn has nowhere to log or suggest improvements.
-
Does AutoLearn automatically add answers to the base?
No. AutoLearn detects gaps and suggests answers, but adding them to the knowledge base always requires your approval. You control what goes in.
-
How quickly does the agent improve with AutoLearn?
It depends on conversation volume and how often you review the panel. In operations with regular traffic, you can see the agent becoming noticeably more accurate within a few weeks.
-
How does AutoLearn avoid cluttering my queue with duplicates?
Semantically similar questions are grouped into a single item. Instead of seeing "How do I export to Excel?" and "Can I download data as Excel?" separately, you see one item with a count of 2 occurrences.
Related features
Knowledge Base
Upload PDFs, links, spreadsheets, and text. SquadOS indexes the content so the agent can search before answering instead of guessing.
AgentMaker
Explain the agent's job in plain language. AgentMaker structures the prompt, model, tools, base, and channels for you to review and publish.
Turn repeated misses into approved content
AutoLearn monitors, groups, and suggests. Your team reviews before publishing to the base.
Start for freeRequires an active knowledge base.