PricingCostArchitecture

The Hidden Cost of Naive Chat History

·5 min read

Why re-sending institutional history as context becomes expensive fast, and how structured memory keeps private-capital AI workflows efficient.

The Hidden Cost

Many AI workflows start with the simplest approach: re-send the entire history on every model call. It works. Until it doesn't.

Here's what happens to your costs over time:

Session 1: You send 1 session worth of context. Cost: $X. Session 10: You send 10 sessions of accumulated history. Cost: $10X. Session 50: You send 50 sessions. Cost: $50X.

But it's worse than linear. Within each session, every exchange also re-sends all previous exchanges in that session. The actual cost function is:

cost(n) = Σ(i=1 to n) [sessions_done × exchanges × tokens_per_exchange × cost_per_token]

This is O(n²) — quadratic growth. At 40 exchanges per session, 5 sessions per day, over 60 days, you're looking at ~$540 in LLM costs for a single user. With Recallr's fixed-overhead pipeline, the same usage costs ~$200.

Why It's Quadratic

The naive approach has a fundamental problem: every new session must include ALL previous sessions as context for the LLM to "remember" anything. Session 50 carries the weight of sessions 1-49.

This means: - Input tokens grow linearly with history length - Cost per session grows linearly with total sessions - Total cost grows quadratically with time

The Recallr Fix

Recallr's cost is linear because the pipeline overhead is fixed per session, regardless of how much history exists:

1. Stage 0 (Context injection): Only relevant memories are injected, not raw history 2. Stage 1 (Session summary): Fixed cost to summarize the current session 3. Stage 2 (Memory extraction): Fixed cost per extracted memory 4. Stage 3 (Graph update): Fixed cost to update the knowledge graph

Total cost per session ≈ $3.33 regardless of whether it's session 1 or session 1,000.

The Breakeven

With default parameters (40 exchanges/session, 5 sessions/day, $3/M input tokens, $15/M output tokens), Recallr breaks even around Day 22. After that, every day saves you money — and the gap widens quadratically.

By Day 60: - Naive approach: ~$540 - Recallr: ~$200 - Savings: 63%

Try It Yourself

The pricing calculator on the homepage lets you plug in your own numbers. The math is open — the simulation code matches our published cost model exactly.

The bottom line: if your workflow keeps re-sending accumulated history, you're already paying more than you need to. And the problem only gets worse with time. See how Recallr's architecture keeps costs linear.