Memory-first setup
Turn on the memory-first workspace, let the onboarding skill set you up, and see what your agents already know.
The default workspace is organised around repositories: specs, rules, findings. The memory-first workspace is organised around two questions instead — what do my agents know? and what are they spending? Same data, different front door.
Alpha
This is opt-in and off by default. The pages it points at are all reachable by URL whatever you choose, so turning it on changes navigation, not access. Turning it off puts everything back.
1. Turn it on
Open Account in the portal and switch on Memory-first navigation.
The preference is stored on your user, not your team, so it follows you across devices and does not change what anyone else sees. The side nav swaps immediately to four sections: Measure, Memory, Guardrails and Team.
Specs, Skills and Automations leave the rail. They have not gone anywhere: a skill is a memory with form=skill and a spec one with form=spec, so Browse’s type filter is where you find them now.
2. Set yourself up
Two commands. Install the skills, then ask your agent to run the onboarding one:
npx -y skills add gjalla/engineering -y
claude "Use the gjalla-onboard skill to set me up."That is the whole setup. The skill installs the CLI, wires the agents you actually use, signs you in, links the repo, and backfills the last week of agent work so there is something to look at immediately.
Use an interactive session rather than claude -p: signing in is the one step that needs you, and print mode has no way to show you the approval code and wait.
3. What you will see
Measure is where you land. It reports only what the telemetry observed directly — sessions, tokens, commits, and the share of tokens going to each kind of work. Because the skill backfills the last week, this is populated on your first visit rather than after a week of waiting.
No dollar figures, deliberately: we cannot price every model exactly, and a wrong total would poison every true number beside it.
Memory → Browse is the corpus: every memory, skill and spec your team has, with facets for type, tag and scope. The first time you open it, gjalla builds the graph from what the team already knows and shows a progress state while it works. That happens once.
Memory → Map is the same corpus as a force-directed graph — similarity edges solid, other relations faint. The fastest way to see clusters and orphans.
Memory → Curate is the review queue: what has not been reviewed, what is disputed, likely duplicates, and anything expired. Suggestions land here for a human to accept or reject; nothing is applied without you.
Turning it off
Switch Memory-first navigation off in Account. The rail returns to the default and every page you were using stays reachable at the same URL.