Customer planning tool

Memory cost calculator

Annual projection · USD

Workload assumptions

Enter the client’s expected monthly usage.

Uses 0.5 support contacts per customer/month and 4.2 customer messages per conversation.

Sources: Corebee contact-ratio benchmark and LoopReply’s 10,000-conversation study. The 30-token query and 120-token response are planning assumptions.

Model
Input $— / 1M Output $— / 1M
Customer activity
users
per month
queries
Conversation size
tokens
tokens
Memory setup
Strategy
Planning case
months
Projected memories / user / year
—
Projected total memories / year
—
Projected memory tokens / year
—
Benchmark tokens / memory
—
Benchmark retrieval latency
—
Benchmark end-to-end latency
—

Estimated ingestion cost

Based on annual source-token volume

Annual memory ingestion
$—
Monthly average
$—

Annual memory ingestion

Cost to process conversations and create searchable memories.

ComponentTokensCost
LLM input—$—
LLM output—$—
Embeddings—$—
Total ingestion$—
Estimated Redis database size —

Based on — new source tokens per month and the selected rolling retention.

Calculations are calibrated to the source data. Memory counts scale with conversation-token volume. Tokens per memory and latency are measured strategy coefficients. Maximum combines independently observed maxima and is an upper-bound planning case.

Prices are list prices per 1M tokens, updated July 2026. Embeddings use text-embedding-3-small at $0.02 / 1M. Query serving, storage charges, caching, discounts, and infrastructure are excluded.

Redis sizing assumes 1,536-dimensional float32 vectors, 25% vector-index overhead, 4 bytes per memory token, 512 bytes of metadata per memory, and 1 KB per session event. The model is calibrated to 36.9 MiB/Instruct and 69.4 MiB/Remis + Instruct per 1M source tokens.