Memory Layer Pricing Math: What 500K Adds Actually Cost

Memory layer pricing usually gets argued at the monthly price. The number that actually drives the bill is the add request — how many memories your agent extracts, not how many API calls it makes. Here is the arithmetic, tier by tier, for solo developers and small teams.

What counts as one add request

Memory layer pricing starts with a counting rule, not a price. Audaxic counts an add request per memory extracted from a message batch. One /add call that extracts five memories counts as five add requests.

That rule explains most billing surprises. Paste a long log into a message batch and the extractor may pull 40 durable facts out of it: that is 40 add requests, not one. Send five short messages that each yield one preference and that is five add requests. The unit you budget in is extracted memories.

Retrieval uses a different unit. Retrieve requests are tiered: the first 10 memories returned count as 1 request, and each additional 10 counts as 1 more, so a search that returns 25 memories counts as 3 retrieve requests. Every plan keeps a 10:1 ratio between adds and retrieves — 50,000 to 5,000 on Starter, 200,000 to 20,000 on Growth — which matches the tier size used for counting.

Tiering means a broad search costs more than a narrow one: 10 memories is 1 request, 100 memories is 10 requests. Focused queries are cheaper to run and easier for an agent to use, so the counting rule pushes in a sensible direction.

A short calculation shows why extraction volume, not call count, moves your bill:

adds per month = memories extracted per session × sessions per day × working days
              = 25 × 3 × 22
              = 1,650

The add endpoint reference covers the request mechanics.

Memory layer pricing, tier by tier

Five published plans, all with unlimited projects and graph memory:

Effective cost per add barely moves across the paid tiers: $0.24 to $0.26 per 1,000 adds. Retrieve quota is always one-tenth of add quota, so the plans scale together — $12 for 50,000 adds and $125 for 500,000 keep the same ratio. In per-retrieve terms: $2.40 per 1,000 on Starter, $2.50 on Growth and Pro, $2.60 on Enterprise. A retrieve request covers up to 10 memories, which is how the 10:1 quota ratio lines up with the tiered counting rule.

If your workload is retrieval-heavy relative to extraction, watch the retrieve column as closely as the adds column.

Hobby costs $0/month, requires no credit card, and still includes 10,000 adds, 1,000 retrieves, unlimited projects, and graph memory. Enterprise adds one option: on-premise deployment available on request. The pricing page has current numbers, and the pricing docs spell out the quota definitions.

Example: a solo developer, three sessions a day

Suppose your workflow extracts 25 memories per session and runs three sessions on a working day. Over 22 working days that is 66 sessions, 1,650 adds, and — at three searches per session — 198 retrieve requests. Both figures fit inside Hobby, so the monthly bill is $0.

Before you estimate, measure. Run a normal week of sessions, then page through what was stored with GET /list and count the memories your workflow produced. Extraction rates vary far more between workflows than between weeks within one workflow, so your own data beats a general assumption about how much a session should remember.

Change the extraction rate and the tier changes with it. A session that digests a migration plan, a style decision, and a dozen gotchas can extract 300 memories. The same 66 sessions then produce 19,800 adds:

Retrieval stays flat at 198 requests because it depends on how often you search, not on how much a session extracts. The adds column drives the tier decision almost entirely. A migration month that extracts 20,000 memories costs $12 on Starter, and the quiet months after it cost the same $12 — the flat-fee tradeoff in miniature. The lighter two profiles never leave Hobby, which is why a solo developer with scoped extraction can run for $0 with no credit card.

Example: a five-person team

Multiply the heavy profile by five. At 19,800 memories per developer the team reaches 99,000 adds, past Starter's 50,000 and inside Growth at $50/month. At 50,000 adds per developer the team needs 250,000, which crosses Growth and lands in Pro at $125.

Every plan includes unlimited projects and graph memory, so adding repositories or teammates does not create a per-project charge. The only lever is monthly extraction volume. Retrieval stays modest here: five developers at 198 requests each is 990 retrieve requests, well under Growth's 20,000 quota.

Growth's 200,000 adds give the 99,000-add team more than double its usage in headroom, so a spike month does not force a move. The next boundary is tighter: 250,000 adds no longer fits Growth and lands on Pro, where 500,000 adds and 50,000 retrieves cost $125. Because quotas are fixed rather than metered, the team's busiest month and quietest month cost the same.

Flat vs overage: why predictability matters

The published comparison as of August 2026 puts the two billing models side by side:

In per-add terms the gap widens with tier: $0.24 versus $0.38 per 1,000 adds at the entry paid tier, $0.25 versus about $0.40 at the middle tier, and $0.25 versus about $0.50 at the top published tier. The absolute differences in the table — $7, $29, $124 per month — are the same numbers expressed the way a budget line sees them.

Matching quotas cost up to 50% less on Audaxic, and billing stays flat: no per-token fees, no overages. The difference matters most in a heavy month. With flat billing, a month where extraction doubles does not produce a variable invoice — you either stay inside your quota or you move up a tier at a published price.

For the full breakdown, read Mem0 Alternative: Cheaper Memory Layer for Coding Agents and the four-way comparison with SuperMemory and Zep.

When to move tiers

The memory layer pricing decision comes down to one estimate and one comparison. Count what your agent extracts, then pick the first tier whose adds quota covers it.

1. Estimate monthly adds: memories extracted per session × sessions per day × working days.
2. Estimate monthly retrieves: searches per session × sessions, plus one request for each additional 10 memories a search returns.
3. Pick the lowest tier that covers both, then recheck at month end. If adds approach the quota, move up.

The boundaries sit at 10,000 / 50,000 / 200,000 / 500,000 / 1,000,000 adds per month. Most solo workflows with structured extraction stay inside Hobby or Starter, and small teams land on Growth or Pro — the examples above show why. Tiers are flat prices rather than per-request surcharges, so the upgrade question is whether the next quota covers this month's extraction, not what an overage would cost. A bulk import of past conversations goes through the same counting rule, so a large import month can cross a tier once; budget for the month that includes the import, not for the average month.

Audaxic exposes a mem0-like API, so if you already know the add, search, list, update, and delete endpoints, the pricing model is the part worth re-checking before you commit. Keep the machine-readable pricing file in version control to notice quota or price changes over time.

FAQ

Written by Efe — Sep 25, 2026 · More articles · X @AudaxicTech