Skills
Memory in Ajeris comes in two flavors. Facts are things the agent knows about you - your address, your wife's birthday, your tone preferences. Skills are things the agent has learned how to do - recipes for handling a specific kind of task that worked the first time and are worth keeping for next time.
This page covers skills.
Why skills exist
The first time you ask the agent to do something complex - "draft my weekly investor update from yesterday's metrics, this week's notable events, and last month's growth chart" - it figures it out from scratch. It calls multiple tools, stitches the results together, and sends a draft.
The second time you ask, you don't want it to start over. You want it to reuse what worked.
Skills make that possible. After a successful run of a complex task, the agent saves a short procedural recipe: what tools to call, in what order, with what arguments, and what pitfalls to avoid. The next time you ask for the same kind of thing, the agent loads the recipe and follows it.
What a skill looks like
A skill is a small structured note the agent writes to itself. It has a name, a one-line description, and a body that reads like a runbook. A typical skill answers four questions:
- When to use it. What kinds of asks should trigger this skill?
- What to do. Step-by-step procedure.
- Common pitfalls. Failure modes to watch out for.
- How to verify. How to know the result is correct.
You don't normally see skills - they live behind the scenes - but you can ask the agent at any time:
- "What skills do you have?"
- "Show me the skill for the weekly investor update."
- "Forget the skill for X."
How skills get created
Two paths:
1. Explicit ask. "Save this as a skill so you don't have to figure it out next time." The agent writes a skill capturing what just worked.
2. Passive capture. When the agent successfully completes a non-trivial workflow (multiple tool calls, recovered from an error, solved a tricky composition), it considers whether the procedure is worth saving for next time. If it is, it saves a skill in the background. You don't see this happen.
Skills are only saved when there's clear signal they're worth keeping. Most one-off tasks don't generate skills.
How skills get used
Whenever the agent thinks about how to handle a request, it sees a short index of available skill names and descriptions. If one looks relevant, it loads the full recipe and follows it. If none apply, it solves the task fresh.
Over time, the index grows. The agent gets noticeably better at the recurring shapes of work in your life - your weekly emails, your end-of-month reporting, your travel-day routine - because it's literally building up procedural memory of your patterns.
Editing and removing skills
You can edit any skill via the agent in plain English:
- "Update the airport-day skill to also remind me to charge my AirPods."
- "The weekly-update skill is including too much detail - shorten the summary to three bullets."
- "Forget the skill for X."
Edits apply immediately. New runs of the relevant task pick up the edited recipe.
Privacy
Skills are stored encrypted at rest, scoped strictly to your account, and never shared between users. See the security policy for the full data-handling details.