შინაარსზე გადასვლა
guide10 ნოემბერი, 2026 · 9 min read

Migrating a crypto exchange to new software: balances, keys, users and the freeze window

How to move a live crypto exchange to new software: what data moves, why keys usually do not, how to reconcile balances, and how long trading has to stop.

BitPolaris Engineering

English original

A crypto exchange migration is the move of a live venue from one software platform to another: users, balances, trade history, wallets and configuration, with trading paused for an announced window while the final copy is taken and checked. The work is not the copy. It is proving that every customer balance on the new system equals the balance on the old one and the coins on the blockchain, and doing it in a window short enough that customers accept it.

This guide is for founders and CTOs who already run an exchange and want to change vendor, move from a hosted instance to their own servers, or replace in-house code. It covers what moves and what does not, why keys are the hard part, the two cutover patterns, the reconciliation that decides go-live, what to put in the contract, and what a migration costs. We do this work as a fixed-price service, and we say where our approach is one choice among several.

Why exchanges change software

The reasons we hear: the vendor changed pricing or direction; a regulator or bank wants the system and keys inside your own perimeter (see the self-hosted guide); revenue share no longer makes sense at your volume; or a piece such as native apps or custody policy is missing.

In the EU there is now a regulatory reason to plan the exit before you need it. The Digital Operational Resilience Act applies to crypto-asset service providers, and its Article 28(8) requires exit strategies for ICT services that support critical or important functions: documented plans that let the firm leave a provider without disrupting its business, its regulatory compliance or the service to its clients. An exchange’s trading and custody software is hard to describe as anything other than critical. A migration plan you have rehearsed is the evidence for that article.

What has to move

Six kinds of data move. They differ in how hard they are and in what you can do if one of them fails.

What What moves Typical problem Fallback
Users and KYC Accounts, verification levels, limits, KYC provider references Old and new systems model tiers differently Map tiers in a table both compliance teams sign
Credentials Email, password hashes, 2FA secrets, devices Hash algorithm or encrypted 2FA secrets cannot be read by the new system Forced password reset, 2FA re-enrolment
Balances Every account, every asset, including fee, hot-wallet and omnibus accounts Rounding, dust, pending deposits and withdrawals in flight None: must reconcile before go-live
History Orders, trades, deposits, withdrawals, adjustments Volume, and different status codes between systems Archive the old history read-only and link to it
Wallets Addresses, keys or key shares, address-to-user mapping Keys cannot be exported from the old custody New keys, new addresses, on-chain sweep
Configuration Markets, fee tiers, limits, liquidity settings Fee rounding and tick sizes differ Recreate and test each market on staging

KYC documents normally stay with your KYC provider; what moves is the reference to each applicant and the decision you recorded. Ask the provider early whether the new integration can use the same applicant records: customers asked to verify again are customers you lose.

Keys: the part that usually does not move

The question that shapes the whole plan is whether you can take the private keys with you. Usually you cannot, and often you should not.

On a hosted platform the keys sit in the vendor’s environment, and a vendor that lets them leave in importable form has a weaker security story. MPC or HSM key shares are bound to a provider or device. Even where export is possible, old keys bring every person who ever had access to them. A new key ceremony with your officers present and a signed record is cleaner evidence for an auditor.

So the usual plan is:

  1. Generate new hot, warm and cold keys on the new system before the freeze.
  2. Issue new deposit addresses to every user on the new system; announce the change in advance and in the app.
  3. During or after the freeze, sweep balances from the old addresses to the new wallets on-chain, chain by chain, with the transactions recorded for the reconciliation file.
  4. Keep watching the old deposit addresses for an agreed period and credit late deposits to the right account. Users will keep sending to saved addresses for weeks; exchanges that skip this step create support tickets and, sometimes, lost funds.

Budget network fees for the sweep, including the cost of activating new addresses on TRON and of moving many small UTXOs on Bitcoin. Our notes on stuck USDT deposits cover the address and network mistakes that also appear during a sweep.

Credentials, sessions and API keys

Customers judge the migration by whether they can log in on Monday. Three items decide that.

Passwords. Hashes in an algorithm the new system supports can be imported as they are. Otherwise, either keep the old hash and rehash on the first successful login (invisible to users, needs code in the login flow) or send every user a reset (simple, and a wave of support tickets).

Two-factor secrets. TOTP secrets are often encrypted with a key that stays in the old system. If you cannot get them in plain form under a controlled procedure, users must re-enrol 2FA after signing in. Make the re-enrolment flow secure, because an attacker who knows a password would like nothing more than a fresh 2FA setup.

API keys. Plan for every API user to create new keys on the new system. Coinbase took this route when it moved its International Exchange onto Deribit on 1 October 2026: institutional API and FIX clients needed new Deribit keys, the old keys stopped working and IP allowlists were not copied across (CryptoSlate, 2 October 2026, read 10 November 2026). Publish the new endpoint documentation and a sandbox before the freeze, so market makers can test before they are cut off.

Open orders should be cancelled at the freeze, not migrated. Coinbase did the same: open orders were cancelled at the halt and clients had to place them again on the new venue (same source). Carrying an order book across two matching engines with different rules creates fills nobody can explain.

Two ways to cut over

There are two patterns, and the choice depends mostly on scale.

One announced freeze. Deposits, withdrawals and trading stop at a time announced in advance. The final export is taken, imported, reconciled, and the domain is switched to the new system. Everyone moves at once. This is the pattern our service uses, and it is what most venues below the very largest choose. The freeze length depends on the number of accounts and chains, so measure it in the rehearsal rather than guessing it in a sales call.

Account by account, behind a replicator. Both systems run in parallel and a replicator keeps the new ledger in sync with the old one while accounts are switched over in small groups. Binance described this approach for its own ledger migration: a dedicated replicator synced balances, two reconciliation modules checked the data (one per transaction, one comparing full snapshots), and each account was frozen for a moment while its ownership moved, about 150 milliseconds on average (Binance blog, “Overcoming Unique Challenges in Our Binance Ledger Migration”, read 10 November 2026). The new ledger went live in early August 2022 and replication ran until mid-November. That is a programme of several months for a venue that could not stop. For a regional exchange, the cost of building and testing a replicator is far higher than the cost of one Sunday-night freeze.

The reconciliation that decides go-live

A migration succeeds or fails on one report. Before the switch, and again after it, every one of these must match:

  • Per account, per asset: the balance on the new ledger equals the balance on the old ledger at the freeze, to the smallest unit the asset supports.
  • System accounts: fee income, hot-wallet, omnibus and adjustment accounts are included; these are where differences hide.
  • Against the chain: the total customer liability per asset is covered by on-chain holdings plus balances held at hedging venues or custodians, with each address listed.
  • In-flight money: deposits seen but not yet confirmed, and withdrawals approved but not yet broadcast, are listed and either completed before the freeze or carried across with their status.

Our rule: if the numbers do not match, we do not go live; every difference is explained and fixed in the rehearsal. Keep the report. It is what you show customers who ask, and what an auditor or regulator will ask for after a material system change. The same evidence that supports proof of reserves supports the migration file.

A rehearsal, then one switch

The rehearsal is a full migration onto a copy of the new system, using a real export from the old one. It produces three things: a reconciliation report you review line by line, a measured duration for each step, and a list of fixes. If the rehearsal fails, you rehearse again. The typical sequence we follow:

  1. Assessment. Export formats, data model, chains, custody setup and the users with API access. The price and date are fixed at the end of it.
  2. Rehearsal. Full migration on a copy, with the reconciliation report.
  3. Announcement. Users are told the date, the expected freeze length, that deposit addresses will change, that API users need new keys and whether passwords or 2FA must be reset.
  4. Freeze and switch. Deposits and trading paused, final export, import, reconciliation, DNS switch, reopen.
  5. Aftercare. Two weeks of close watch, late deposits on old addresses credited, the old system kept read-only for reference.

Keep users on the same domain, so the switch is a DNS change, and keep old history readable, migrated or in a read-only archive. Coinbase kept historical International Exchange trade and order data available through the old APIs for about 12 months after the move (CryptoSlate, as above).

What to put in the contract with your current vendor

Much of the risk is fixed in the contract you signed at launch. With any vendor, including us, check:

  • Exit notice and data export. How much notice, in what format, and does the export include balances, history, KYC references and configuration, or only users? On our cloud model the terms are 90 days’ notice and a full data export.
  • Key handling at exit. Will the vendor sign sweep transactions during the exit period, and within what time?
  • Read-only access afterwards. How long can you reach the old admin panel and logs for audits and disputes?
  • Destruction. Will the vendor destroy its copy of your data and confirm it in writing? When customers move from our cloud to their own servers, the cloud instance is kept read-only for 30 days and then destroyed with a certificate.

What a migration costs

Our migration service is fixed-price after the assessment, and the price does not change if the switch takes longer than planned:

Scope Typical case Price
Standard One exchange, standard export, up to 10 chains from €9,500
Larger Custom or in-house system, more chains, several entities €15,000–25,000
Licence BitPolaris on cloud, self-hosted or source code as in the price list

For the licence, the exchange block is €24,500 one-time self-hosted plus 20% a year for updates and support, or €7,700 setup plus €1,600 a month on the cloud model. Your own costs come on top: network fees for the sweep, support staffing for the week after the switch, any KYC provider charges for a new integration, and the time of your compliance team to sign the tier mapping and the reconciliation. If you are comparing destinations, our vendor comparison lists who publishes prices and who hosts the keys.

Crypto exchange migration FAQ

How long does trading have to stop during an exchange migration?

It depends on the number of accounts and chains, and on whether credentials can be imported. With one announced freeze, measure the duration in a rehearsal on a copy and announce it to users in advance. Coinbase’s move of its International Exchange onto Deribit was completed in one scheduled maintenance window on 1 October 2026; very large venues sometimes move account by account to avoid a freeze, at much higher engineering cost.

Can I move my private keys to the new exchange software?

Usually not, and it is often better not to. Hosted platforms rarely let keys leave their environment, and MPC or HSM key shares are bound to a provider. The standard plan is new keys from a new key ceremony, new deposit addresses for users, an on-chain sweep of balances and monitoring of old addresses for late deposits.

Will my users have to reset their passwords and 2FA?

Only if the new system cannot read what the old one stored. Password hashes in a supported algorithm can be imported, or rehashed on first login; otherwise users get a reset link. TOTP secrets that are encrypted inside the old system cannot be read, so those users re-enrol 2FA. API users always need new keys.

What happens to deposits sent to old addresses after the switch?

They should still be credited. Keep the old deposit addresses monitored for an agreed period, sweep what arrives and credit it to the right account. Tell users in advance that addresses change, and show the new address in the app before the freeze.

What does it cost to migrate a crypto exchange to new software?

At BitPolaris the migration itself is fixed after an assessment: from €9,500 for one exchange with a standard export and up to 10 chains, and €15,000–25,000 for in-house systems or several entities. The software licence is separate and follows the public price list; network fees for the sweep and your own staff time come on top.

BitPolaris moves exchanges onto its software from hosted platforms, open-source kits and in-house code, with a rehearsal on a copy and a balance reconciliation before and after the switch. The exchange runs on our cloud, your servers or from source.

Exchange migration service · See the Exchange block · Pricing

guide · 6 წუთის საკითხავიSelf-hosted კრიპტობირჟის პროგრამული უზრუნველყოფა: რას იღებთ, რას მართავთ და რა ღირსრას მოიცავს self-hosted კრიპტობირჟის პროგრამა, რა უნდა მართოს თქვენმა გუნდმა, რით განსხვავდება white-label ღრუბლისა და ღია კოდის ნაკრებისგან და რა ღირს.cost comparison · 9 წუთის საკითხავიკრიპტობირჟის პროგრამის მომწოდებლების შედარება (2026): განთავსება, საჯარო ფასები, ლიკვიდობა და გასაღებებიკრიპტობირჟის პროგრამის 14 მომწოდებელი გვერდიგვერდ, მათივე გვერდებიდან 2026 წლის 1 ნოემბერს: განთავსება, საჯარო ფასები, ლიკვიდობა, გასაღებები, აპები, ვადები.operations · 11 min readWhy USDT deposits get stuck: the failure classes we found in productionSilent minimums, sweeps marked done without a receipt, an empty gas tank, scanners that die quietly: deposit failures we found running the reference exchange, and the alert for each.

დაჯავშნეთ 30-წუთიანი შეფასების ზარი

Crypto exchange, payments, wallet, custody and token launch software, already running in production. Run it on your own servers, license the source code, or let us host it under your brand.