The knowledge base

Every investigation adds to your team's shared knowledge base, so a known issue gets a known answer and what one engineer works out helps everyone.

In most robotics teams, the answer to last month's failure lives in one engineer's head. When the failure comes back, someone starts again, or waits for that engineer.

Alloy keeps that knowledge for the whole team. Every investigation, every support case and every correction adds to a shared knowledge base. The next time the issue appears, the agent starts from what worked. Solve it once, and the whole team has the answer.

What the knowledge base keeps

What	Where it comes from
Known issues and the fixes that worked	Each support request becomes a case with its outcome. The agent captures the outcome, including with a follow-up when a thread goes quiet.
Who to escalate to	The agent learns which person or team resolves which kind of issue.
Notes your team asks it to remember	"Remember that…" in chat, for you or your whole organisation.
Your trusted methods	Scripts and documents in the Scripts library. The agent runs your method, not a new one.
Your guides and runbooks	Your Confluence triage guides and runbooks, when you connect them.
Past reports and comments	Every report and every comment on a mission stays available to the agent.

The agent says when it uses something it kept, so your team can see why it gave an answer.

It is written by the work, so it does not go stale

Most knowledge bases fail the same way: someone has to write them, nobody has time, and they go out of date. Alloy's is different.

The work writes it. Cases, outcomes and corrections go in as your team works. Nobody sits down to write documentation.
It updates when it is wrong. Correct the agent in a thread and ask it to remember. The next answer uses the correction.
Your team can read it. The knowledge base renders as readable pages for each support role. Anyone in your organisation can check what the agent knows, and Alloy edits it with you.
It sits next to the data. Every note can point back to the cases and recordings it came from, so a teammate can check the evidence before trusting it.

What changes for your team

A known issue gets a known answer. The second time a fault appears, the agent recognises it and gives the fix that worked, with the evidence. No new investigation.
New engineers start with the senior engineer's method. On day one, a new hire can ask the agent and get the answer the team's most experienced engineer would give.
Knowledge stays when people move. When an engineer changes role or leaves, what they solved stays with the team.
Field staff stop waiting on experts. Support and field technicians get the team's past fixes in the thread, instead of calling the one person who remembers.

"Our rate of duplicate tickets has gone down a fair bit, and the mean time to resolution for previously seen bugs has gone down."

Tom Cameron, Tech Lead, AI Engineering. Andromeda

Example

Fictional example: the company, robots and numbers are invented.

A cleaning robot at a hospital site stops with a brush over-current fault. In March, an engineer traced the same fault to a cable tie caught in the spindle, and the case was closed with that fix.

In June the fault comes back at a different site. The agent finds the March case, checks that the motor current profile matches, and replies in the support thread in minutes: the likely cause, the fix that worked last time, and a link to both recordings. The field technician clears the spindle. Nobody pulls an engineer off the roadmap.

Personal memory compared with a team knowledge base

	An engineer's own AI setup	Alloy's knowledge base
Who it helps	The engineer who built it	Everyone in the organisation
Where it lives	One laptop	Your Alloy workspace
Who keeps it current	That engineer, when they have time	The work itself, and Alloy with you
Can the team check it	No	Yes, as readable pages linked to evidence
When the engineer leaves	It leaves too	It stays

How to start

Alloy switches the knowledge base on with you when you set up a support role. It starts filling from the first case. To give it a head start:

Connect your guides. Point the Confluence connection at the triage guides your team trusts.
Save your scripts. Add the analysis scripts your team signs off in the Scripts library.
Turn on a support role. Each request in the channel or Jira project becomes a case with its outcome. See Support Agent.
Correct it out loud. When the agent is wrong, say so in the thread and ask it to remember.

More detail: Teach the agent your team's context.