Live Summaries — what each lim is doing right now
Instead of a static "researching" badge, each lim shows a live 3-6 word status line that reflects their actual current activity. Refreshed every 30 seconds.
What they look like
On the Dashboard, under each lim's name and state badge:
● Searching TechCrunch for Claude news● Drafting Telegram post about GPT-5● Reading carousel_slides.json
The green dot pulses — signaling it's live and updating.
How they're generated
A background service runs every 30 seconds. For each active lim (not idle), it:
- Pulls the last 5 tool calls from the lim's activity trail (capped to keep it cheap)
- Sends a short prompt to the LLM: "In 3-6 words, present tense, what is this lim doing?"
- Stores the answer in the lim's record
- Dashboard displays it next to the state badge
Why this matters
- Honesty — you see what's actually happening, not a generic label
- Progress confidence — lims that are taking a while feel alive instead of frozen
- Problem detection — if Aria's summary stays "Scanning TechCrunch for Claude news" for 5 minutes, you know she's stuck and can intervene
- Character, not mechanic — each lim feels like a real working teammate
Cost
Each summary is ~250 input tokens + ~20 output tokens. With 3 active lims firing every 30s, that's roughly 30k tokens/hour — negligible on your own model (Spark) and affordable on hosted providers.
Turn it off
If you want to save tokens or reduce traffic, set the env var LIMAZE_LIVE_SUMMARIES=0 on cloud-api. Lims revert to showing their current task's first 120 characters instead.