Self-improvement

gaia is meant to grow day by day. The self-improve loop looks back over how you’ve used it and proposes concrete improvements — a new skill for a task you repeat, a sharper soul, or a durable fact worth remembering.

How it works

gaia grow run runs an analyst over recent usage and produces an analysis report of typed proposals — skill, soul, and memory changes. Applying a report is versioned with git (under gaia’s state), so every change gaia makes to itself is reviewable and revertible.

Two modes (set in the analysis config block):

• Autonomous — gaia applies its own proposals (each git-committed). The daemon can also run a cycle periodically.
• Review (HITL) — proposals are queued for you to approve first (analysis.autonomous = false).

Run + review

gaia grow run                  # analyze recent usage → proposals (model key needed)
gaia grow list                 # what gaia has changed about itself (from git history)

In chat, /grow shows the same self-change history. Full flags: Reference → CLI.

Code map

Concern	Module
The analyst + proposal types (skill/soul/memory)	src/gaia/analysis/analyst.py
Applying a report, git-versioned	src/gaia/analysis/apply.py
gaia grow CLI / /grow command	src/gaia/cli/grow.py, src/gaia/commands/grow.py

Config: the analysis block — see Reference → Config.