Article
How Bella memory and context limits work by plan
Last updated: June 22, 2026 · 3 min read
Bella can remember agency facts, preferences, and working rules across sessions. Memory is useful, but unlimited memory in every AI request creates cost, noise, and slower responses. Bella now uses plan-based memory controls.
Two controls work together
- Per-turn context limit: Bella caps how many facts, episodes, entities, insights, and recent turns are injected into the prompt.
- Stored-memory limit: Bella checks tenant-scoped memory count before saving new long-term memory.
- If the account is at the plan limit, Bella skips the new memory write before spending embedding tokens.
What happens near the limit
Bella keeps answering. The chat does not shut down just because memory is near or at the limit. Instead, Bella uses the best available context, skips extra memory when needed, and returns an upgrade or add-on signal so the product can show the right prompt.
Why this matters
- Starter accounts stay lightweight.
- Professional and agency accounts get more useful recall.
- AI budget is protected from unlimited context injection.
- Answers stay focused instead of drowning in old context.
Tenant isolation
Memory counts are agency scoped. One agency's stored-memory usage cannot block another agency, and memory search/write paths remain filtered by the caller's agency.