Pay only for what you use.
Wallet-based metered pricing across the entire substrate -- Records, KG, Causal, Analogy, and the Pinecone-compat memory layer -- one ledger, sub-cent precision, no per-namespace fees, no per-seat fees, no minimums.
How it works
- Sign up. You get $5 of free credit.
- Top up your wallet via PayPal -- $10, $25, $100, $500, or any custom amount.
- Use the API. Every operation is metered and deducted from your wallet in real time. The dashboard shows the per-op ledger.
- We email you when your balance drops below $5. Auto-topup optional.
- Cancel anytime. Any unused balance is refundable for 30 days.
Unit rates
| Operation | Rate | What you'd otherwise pay |
|---|---|---|
| Records -- the primary substrate | ||
| records_ingest (typed event + auto-embed) | $2.00 / 1M | Mem0 Pro caps at 500k records / mo for $249 Zep $125-375/mo flat tiers |
| records_query (semantic + typed filters) | $2.00 / 1M | LangSmith $500 / 1M traces (capture-only) Pinecone serverless ~$17 / 1M reads |
| records_timeline / records_get | $0.50 / 1M | -- |
| records_forget / records_delete | Free | GDPR ops should never be metered. |
| .neruva export / import (atomic backup) | 10 / mo free, then $0.10 / op | No vendor offers atomic single-file substrate export. |
| HD substrate -- nobody else offers any of these | ||
| HD KG add / query / stats | $0.50 / 1M | not offered |
| HD analogy (A:B::C:?) | $0.50 / 1M | not offered |
| HD causal (observe / intervene) | $1.00 / 1M | not offered |
| Pinecone-compat memory -- migration on-ramp | ||
| memory_upsert / memory_query | $2.00 / 1M | Pinecone $4 / 1M write, $17 / 1M read |
| memory_fetch / memory_update | $0.50 / 1M | Pinecone $2-4 / 1M |
| memory_delete / memory_forget_where | Free | -- |
| Storage | ||
| Records / vectors (1-bit packed) | $0.10 / GB / mo | Pinecone $0.33 / GB / mo Weaviate $0.255 / GiB / mo Chroma $0.33 / GiB / mo |
| Egress | Free | Chroma charges $0.09 / GiB egress |
No per-namespace fees. No per-seat fees. No per-pod fees. No minimums. The HD substrate operations replace LLM calls for the substrate-answerable slice of agent workloads, so the relevant comparison there is per-1M-token cost on a frontier model -- which runs orders of magnitude higher.
What you'd pay for the equivalent stack
A Series-A startup with 100 production agents typically wires up three tools: a vector DB for memory, an observability platform for captured turns, and a cross-session memory product. We replace all three. Workload assumed: 500k records/day, 2M queries/day, 5GB storage.
| Layer | Vendor | Monthly |
|---|---|---|
| Vector memory | Pinecone Serverless | ~$1,087 |
| Observability / capture | LangSmith Plus + per-trace | ~$30,112 |
| Cross-session memory | Mem0 Pro (cap exceeded; Enterprise needed) | $249+ |
| Equivalent stack | three vendors, three bills, no KG / causal / analogy | ~$31,448 / mo |
| Neruva | one vendor, one wallet, all four substrate layers | ~$150 / mo |
Sources: pinecone.io/pricing, langchain.com/pricing, mem0.ai/pricing. Swapping LangSmith for Helicone Team ($799 flat) drops the bundle to ~$2,135/mo -- still 14x our rate, and you lose per-trace fidelity. The bundle does not include KG, causal, or analogy -- those have no comparable in any other product, at any price.
Plus: stop burning input tokens on context-stuffing.
Most agent stacks "remember" by re-prepending recall text to every LLM call. That recall slice gets billed at frontier-model input rates, every turn, forever. Replace it with one records_query.
records_query with typed filters. Returns the same 5 KB.An HD causal do(X) query at $1 / 1M replaces a multi-step LLM reasoning chain (~5k input + 1k output ≈ $0.05 / call on Opus 4.7). That's ~50,000x cheaper for substrate-answerable questions -- and you gain determinism, replayability, and no temperature noise.
Deposit cost ≈ 1 call. Recall cost ≈ 1 call. At $2 / 1M ops, that's effectively $0.000004 for a write-then-read round trip. A single high-quality mistake-record can save 15-60 minutes of re-debugging the next time the agent encounters the same root cause. The discipline is one line: deposit at every seam, compact when noise creeps in. The substrate doesn't fire in the inner write--run--fix loop -- it fires at the boundaries where context would otherwise be lost.
What you actually pay
10k records/day, 50k queries/day, 100MB storage. One autonomous coding agent running 24/7. Fits inside the $5/mo free credit.
500k records/day, 2M queries/day, 5GB storage. 100 agentic users in production. ~200x cheaper than the equivalent Pinecone + LangSmith + Mem0 stack.
10M records/day, 40M queries/day, 200GB storage. Equivalent three-vendor stack: ~$60-380k/yr depending on observability tier.
Always free
Enterprise
Volume discounts at $1k/mo committed (-20%), $5k/mo (-40%), and $20k/mo (-60%). For dedicated capacity, VPC peering, SOC 2 audit support, or on-prem deployment, contact info@neruva.io.