MemMesh — persistent, self-improving memory for AI agents. Get started →
Self-HostingLicensing & Caps

Licensing & Caps

MemMesh enforces capacity and feature access through a license that maps to a plan. Plans are sized along three dimensions, each with a distinct role.

The three dimensions

  • Subjects — the headline ladder (stock limit). A subject is one active canonical entity in the memory graph. This is the number a buyer reasons about; it’s enforced by a live count of active entities, not a meter.
  • Predictions — the value meter (flow limit + overage). Bundled per plan; on the higher paid tiers, usage beyond the bundle bills as overage.
  • Events — fair use (flow limit, no charge). A generous monthly cap that hard-stops with an upgrade prompt when exceeded — never an automatic bill.

Plan tiers

TierPrice/moSubjectsProjectsPredictions/moEvents fair-useRetention
Free$01,00011,00010,00030d
Starter$195,00015,00050,00090d
Growth$7925,000320,000200,000180d
Pro$249100,000Unlimited50,000500,000365d
EnterpriseCustomUnlimitedUnlimitedUnlimitedUnlimited7 years

Higher tiers add capabilities such as compliance tooling, SSO, vertical estimators, VPC deployment, and on-device licensing. Yearly billing is offered at 10× monthly (two months free).

Local Free tier vs. the plan number. The plan table’s Free row is sized at 1,000 subjects (active canonical entities). The single-binary local engine enforces a separate, simpler cap: 500 items — a raw count of stored memory rows, set by entry_cap = 500 under [free_tier] in ~/.thinkfleet-memory/config.toml. They measure different things (memory items vs. distinct subjects), so don’t conflate the two numbers.

Entitlements

Beyond the numeric limits, the plan attaches feature entitlements — booleans that gate capabilities like compliance export/erasure, SSO, the health/finance verticals, on-device activation, and VPC deployment. A feature that isn’t attached is denied; an attached limit that’s left unset means unlimited.

Fail-soft behavior

When the licensing/billing backend is unreachable, numeric limits fall back to the Free-tier defaults and feature booleans default off — fail-soft, never wide-open.

This page summarizes the plan model for self-hosters. Exact entitlement codes, overage pricing, and enforcement details are maintained in the engine’s billing configuration and may change — treat the figures here as indicative.