Sessions

Sessions bootstrap your AI with relevant context at the start of a conversation and track skill effectiveness over time. Instead of re-explaining your preferences and workflows every time, start a session and let Kyew bring your AI up to speed.

What is a Session?

A session is a structured way to load your AI's context for a specific domain of work. When you start a session, Kyew returns:

• Recent memories relevant to the domain
• Active skills that apply to the work you're about to do
• Detected patterns worth reviewing
• Stats on how your skills have been performing

This gives your AI a running start — it already knows your preferences, workflows, and what's worked before.

Starting a Session

At the beginning of a conversation, tell your AI what you're working on:

"start a session for client reporting"

Behind the scenes:

session(action="start", domain="client-reporting", context="Preparing this week's client report")

Your AI receives all the relevant context it needs to help you effectively.

Options

• memory_limit — maximum number of memories to return (default varies)
• skill_limit — maximum number of skills to return
• format — "detailed" or "compact"
• skill_detail — level of detail for skills: "full" or "summary"
• summarize — true for an AI-generated summary of the context

session(action="start",
  domain="client-reporting",
  context="Weekly report prep",
  memory_limit=10,
  skill_limit=5,
  format="detailed")

Finding Knowledge

During a conversation, use apply to search for relevant memories and skills for a specific task:

"what do I know about formatting quarterly metrics?"

session(action="apply", task="formatting quarterly metrics", domain="client-reporting")

This returns ranked suggestions from your memories and skills that are most relevant to the task at hand. Use the optional limit parameter to control how many results come back:

session(action="apply", task="setting up a new client onboarding", domain="client-reporting", limit=5)

Reporting Outcomes

After using a skill, report whether it helped:

"that reporting skill worked perfectly"

session(action="reflect", skill_id="skill_xyz", outcome="success", notes="Template was exactly right, saved 20 minutes")

For skills that need improvement:

session(action="reflect", skill_id="skill_xyz", outcome="failure", notes="Template was outdated, needed the new Q2 format")

Outcome tracking feeds back into Kyew's pattern detection. Skills that consistently succeed get higher confidence. Skills that fail can be flagged for revision.

Best Practices

Start Sessions Early

Begin every conversation with a session start. Your AI performs better when it has context from the outset rather than discovering it mid-conversation.

# First message in any work conversation
"start a session for [domain]"

Use Apply Before Complex Tasks

Before diving into multi-step work, check what knowledge is available:

"apply my knowledge about deploying to production"

This surfaces relevant skills and memories you may have forgotten about.

Always Reflect After Using a Skill

Feedback is what makes Kyew improve over time. Even a quick note helps:

"that skill worked" or "that skill was outdated"

Without reflection, Kyew can't distinguish effective skills from stale ones.

API Reference

For the full technical reference including all parameters and options, see session Tool Reference.