Five Things to Check When Choosing an RPC Provider

Published
November 4, 2026
Share

‍

1. Who runs it, and how you know

Start with the basic question: who actually operates the nodes? Some providers run their own bare-metal or cloud fleets; others resell another provider’s endpoints under their own brand. Neither is wrong, but it changes who is on call when something breaks, where your requests are processed, and whose outage you inherit. Ask which client software runs on each chain, in which regions, and whether the same operator also runs the archive tier.

A good provider names its operating company, publishes a security contact, and says plainly what it runs. If a vendor will not describe its architecture, you cannot judge its failure modes. If it is a reseller, your reliability is its upstream’s reliability, and you should ask for that upstream’s evidence too. Ask for the answers in writing, not on a sales call.

Then check that the claims can be tested. A status page with history, a documented limits table and a named contact are all things you can verify yourself. Read the public status page and the contract terms side by side, and compare both with what the marketing site says.

‍

2. Reliability evidence

Every provider publishes a high availability number. What matters is how it is measured and by whom. A figure computed from the provider’s own load balancer logs cannot see an outage that the load balancer itself caused. Look for availability measured by independent probes from several regions, a stated definition of what counts as a down minute, and a history that includes the bad months as well as the good ones.

Incident reviews are the second signal. A provider that writes up outages with a cause, a timeline and a fix is showing you how it behaves under pressure. Ask how quickly it detects chain-level problems such as a client bug or a halted chain, and whether it tells customers before they notice. A long service history with no published incidents usually means nothing is being published.

Look at head lag as well as uptime. Head lag is the number of blocks a node trails the canonical chain, and an endpoint can be up and still return stale data. A provider that publishes head lag next to availability is measuring the thing your application actually depends on.

‍

Measure it yourself

Before committing, run a trial from the regions your users and servers are in. Send a realistic mix of calls for a week, record error rates and latency percentiles, and compare the provider’s numbers with your own. Free tiers exist for this purpose; use them with your real access patterns rather than a synthetic benchmark.

  1. 3. Latency, cached and uncached

Latency depends on where you call from, so test in your own regions. Separate two cases that vendors often blend together: cached and uncached reads. A read of immutable data, such as a block by hash, can be served from a cache and be very fast. A call that must run against current state, such as eth_call at the latest block, cannot be cached for long and shows the real speed of the node behind it. Ask for percentiles (p95 and p99, not averages) for each category in the region nearest you, and ask what happens at the tail when a backend is slow.

  1. 4. Limits

Limits decide whether your workload runs at all, and they are often buried. Check the eth_getLogs block range and result cap, the maximum JSON-RPC batch size, the maximum response size, the number of WebSocket subscriptions per key, and whether archive and trace access (debug_ and trace_ methods) is included, weighted or unavailable on your plan. Ask how rate limiting works and what a client sees when it hits a limit, such as HTTP 429 responses, and whether short bursts are tolerated or cut off.

  1. 5. Security, data and terms

Ask how the provider protects your credentials and your traffic. API keys should be stored hashed, shown once, rotatable without downtime and scoped per project. Ask what is logged, for how long, and whether full IP addresses are retained. Ask for the assurance programme in specific terms: dependency scans, penetration tests, who performs them and how often, and a vulnerability disclosure policy with a stated response time. Treat any claim that is not tied to a standard or a date with caution. If a SOC 2 report is not yet available, ask for the audit window dates. Finally, no RPC provider should ever hold your private keys: an endpoint receives signed transactions, never keys.

  1. Data handling

Data handling matters even for public chain data, because your request pattern is itself sensitive. Ask where requests are processed, whether logs contain wallet addresses or full IP addresses, what the retention periods are, and who the sub-processors are. If you operate under Australian privacy law or an APRA standard, ask whether the provider will sign a data processing agreement and supply evidence for your own supplier assessment.

  1. Support and contract terms

Read the contract as carefully as the datasheet. Check how credits work, whether they are cash or service credit, the cap, the claim window, and when you may leave if the provider misses its commitment repeatedly. Ask who your named contacts are, what response time applies to a severity 1 incident, and whether a root-cause report follows. Confirm governing law, the liability cap and the data processing terms before you sign, not after.

Finally, assume that any single provider will have a bad day. Design your client with a fallback: a second provider or your own node behind a failover policy, retries with backoff for idempotent reads, and timeouts shorter than your users’ patience. Keep provider URLs and keys in configuration, test failover before you need it, and avoid depending on provider-specific extensions unless you have an exit plan. Switching cost is part of the price you pay.

‍

How Tessellate answers

Tessellate is new, so we cannot show you a long track record; we can show you a design and targets you can check. RPC launches on six chains in Q1 2027, and we publish each target against measured results on our status page, per region and per month, including the months we miss. Availability is measured at our edge: 99.9% on shared tiers, p95 under 60 ms for cached reads and under 250 ms for eth_call in the nearest region, and head lag within two blocks of canonical for 99.9% of minutes. These are targets, not history.

Who runs it: Tessellate Labs Pty Ltd, Sydney, on bare-metal node pools in colocation, with client diversity per chain (on Ethereum, Reth with Lighthouse and Geth with Prysm). Limits are published per tier: eth_getLogs ranges of 2,000, 10,000 or a configurable default of 50,000 blocks, with a Reth archive tier for history and trace methods. Judge us on the status page rather than on this article, because we are new and have no history to point to.

Security and terms: we never hold customer keys or assets, and API keys are hashed at rest and shown once. Enterprise contracts run under NSW law with an MSA, DPA and SLA schedule; Severity 1 response is 15 minutes with a root-cause report within five business days. 24x7 Severity 1 cover starts in Q3 2027, once the on-call rota has four engineers in two time zones.

Questions about any of this? Email hello@thetessellatelabs.com or read the documentation at docs.thetessellatelabs.com.

‍

Further reading:

Reorgs, finality and what latest really means

Ten years of Ethereum: lessons for node operators

Australian AML/CTF reform and blockchain infrastructure

Tessellate status page

‍

About Tessellate

Tessellate Labs Pty Ltd is a blockchain infrastructure company based in Sydney, Australia. We build RPC and indexing across EVM chains, plus smart contract audit tooling, and we publish uptime and latency against targets on a public status page.

The information herein is provided for general information only. It is not legal, financial or investment advice, and targets described here are targets, not guarantees. Commitments apply only as set out in a signed agreement.

Explore More From Tessellate

Bring reliable infrastructure to your organisation

Meet with us

Tessellate respects your privacy. By submitting this form, you are acknowledging that you have read and agree to our Privacy Policy, which details how we collect and use your information.