Service knowledge that updates from production
Each service has a living wiki that updates after every change and investigation.
Every agent starts from what your team already learned.
Agent memory holds what NOFire AI learns from production, the learnings your team approves and your runbooks.
Learnings a person approves
When an investigation closes, a person approves or skips each learning.
Your runbooks and post-mortems
Add runbooks and post-mortems, tied to the services they cover.
Recall with a citation
Agents load memory with a tool call, and every answer lists the knowledge used.
What memory for agents takes
Memory that agents can rely on needs review, scope, freshness and citations.
NeedBuild in-houseWith NOFire AI
SourcesWikis, chat threads and post-mortems in separate toolsLearned, approved and written knowledge in one memory
ReviewAnything the agent writes is savedA person approves or skips each learning
ScopeOne shared pile of notesEach entry tied to the services the entry covers
FreshnessMemory goes stale when production changesMemory follows production, and outdated learnings decay
CitationsNo record of what the agent readEvery answer lists the knowledge used
CorrectionsWrong entries stay until someone finds themEdit or hide an entry, and the next run reads the change
Cost of ownershipA team to build retrieval, review and scopingIncluded in the platform
Frequently asked questions.
Where does memory come from?
Three sources: service knowledge NOFire AI learns from your production environment, incident learnings a person approves when an investigation closes, and the runbooks, post-mortems and playbooks your team adds.
Who approves what agents learn?
A person, when the investigation closes. Each proposed item can be edited, approved or skipped.
Can we bring our existing runbooks and post-mortems?
Yes. Add each document as a knowledge entry and tie the entry to the services the document covers.
How do agents use memory?
Agents read memory through tools before they answer or act, and every answer lists the knowledge used.
How do we fix a wrong entry?
Edit the entry, or hide a learned entry. The next run reads the change.