How lims & the studio self-learn
Each lim carries persistent memory across sessions. The studio itself accumulates lessons. Nothing is lost between conversations — lims genuinely become themselves over time.
Two layers of memory
1. Per-lim memory
Every lim has a profile with:
- Soul — core personality (given by template or training)
- Identity — self-description
- Voice — how they write / speak
- Operating rules — do's and don'ts earned through work
- Strengths & constraints — what they excel at and avoid
- Custom instructions — user-authored directives
- Memory summary — rolling digest of what they've done
Each run injects this memory into the lim's prompt — so Aria today is the Aria who learned last week.
2. Studio-level memory (Nexus's memory)
Nexus holds the studio's collective memory:
- Preferences — user stylistic choices observed
- Constraints — limits discovered through trial
- Learnings — patterns observed across runs
- Mission statements — strategic directions you've signaled
- Policies — risk tolerance, approval strictness, publish aggressiveness, client-facing tone
Automatic lesson extraction
After every successful orchestration, the cloud-api runs an extractor:
- Takes the user request + plan + step outcomes
- Asks the LLM: "What 1-2 durable lessons should Nexus remember?"
- Classifies each as preference / constraint / learning / mission
- Writes them to Nexus's memory (max 40 notes, oldest pruned)
The lessons are durable ("user prefers 3-sentence Telegram posts", "Marcus is slow at carousels") and specific — not generic AI advice.
How memory shapes future decisions
- Next time you ask Nexus to orchestrate similar work, the past lessons are injected into his system prompt
- Nexus references them implicitly — "keeping this short per your pattern last week"
- Lims inherit studio-level voice & policies through their own system prompts
- No need to re-explain preferences every session
Inspecting memory
Studio Settings → Memory shows:
- Every memory note Nexus has recorded (kind, summary, detail, when)
- Policy toggles (risk tolerance, approval strictness, etc.)
- Option to manually pin or delete notes
Privacy note
Memory is scoped to your studio — no cross-studio leakage. Tenants sharing your studio via marketplace see only what their role permits (read-only on lim profiles; no access to policies by default).