Skip to main content


The CVM is made of opaque contracts for execution of non-atomic, asynchronous, trustless, transparent, non-custodial, bridge-agnostic, generalized and Turing decidable language cross-chain programs and intents.

Why CVM?

To enable users to perform cross-chain intent settlement such as transferring DOT to Osmosis, swapping to ETH, transferring ETH to Ethereum, swapping to USDC, and moving funds to Composable Cosmos or any desired destination chain, in one user signed transaction. Users should be able to express in simple english their desired actions.

Here is what should happen semantically:

This sounds simple, but you need to know several encodings, transports protocols, and SDKs to do this. Also, without the CVM, one may not easily start on Polkadot and proceed to the end using a single transaction.

CVM solves it this way (displayed in a simplified manner):

More detail on this architecture can be found in the rest of this documentation. Carrying out cross-chain swaps is just one of the many possible operations


  • Assets, tokens, coins, funds, amounts: used interchangeably
  • Chain, blockchain, consensus, domain: also used interchangeably to refer to execution systems using different instances of proofs systems
  • Address, account, wallet: also used interchangeably; each can do something on-chain from its own identity, and hold some funds
  • Bridge, relayer: Something which can act on one chain on behalf of another, for example mint tokens (transfer)

Definition deconstruction

Opaque contracts - In the CVM, these may be module or contract, executing on one or many chains

Non-atomic - the whole program is not executed in single block; it is not an all-or-nothing transaction

Asynchronous - two Spawns in single program, but are sent to bridges at same time, execute concurrently on on several chains; other instruction are sequential

Trustless - internally, the CVM uses the most secured, trustless bridges, to execute CVM programs; CVM contracts are subject to cross-chain governance

Non-custodial - each user owns all assets on each chain; they delegate the ability to move these funds cross-chain from their cross-chain account (executor)

Turing decidable language - that there are no procedural loops or recursion, meaning that CVM programs are predictable, deterministic, and will stop

Transparent - reading and auditing CVM program much easier than set of cross chain contracts, code of propagated along with data, you do not need to deploy programs, you just send programs; as subprograms move from chain to chain, they move assets along

Bridge-agnostic, generalized - CVM abstracts away several underlying transports,encodings, assets systems, and messaging protocols, you do not need to know assets systems, how to talk to bridges, or how to encode common operation on each chain you want to act on; you learn the CVM, and it helps you to handle these details

Cross-chain programs - the same program input works consistently on all supported chains; you can write it once to run anywhere

Accessing de facto native liquidity markets - the CVM uses native assets contracts on each chain and each bridge it uses, so programs can leverage existing liquidity on chains; as new bridges and chains are added, the CVM incorporates these without any modification from the user

Additional References

Cross-chain DeFi gets complicated. To understand this documentation at the deepest level, it is helpful to have existing knowledge about the application side of messaging protocols, like Parity’s XCM, Cosmos’s IBC-ICS-20/27, or NEAR’s multi block execution. Also has some resemblance to Fluence Aqua and Anom Juvix

What is CVM, and what are its functions?

The following aims to answer this question. You can being reading any of these sections first, perhaps switching back and forth between sections several times as needed; the contents of these categories partially overlap

Building blocks

The CVM is a set of opaque contracts on chains. Each opaque contract may be one contract on a specific chain or several, or it can be a module. We will refer to this simply as a contract from this point onwards.

A CVM program is a tree of instructions with attached assets. Each opaque contract has a program as input. The program format is an executable language to describe what the user wants to do across chains.

Instructions inside programs are interpreted by the Executor. It owns all of the user's funds. The executor delegates bridges to manage funds operations cross-chain. The executor is the user's cross-chain account.

Instructions include, but are not limited to:

  • Transfer: transfer funds from account to account on a single chain
  • Spawn: transfer leaf of tree (subprogram) to other chain with assets, for further execution
  • Call: arbitrary ABI builder call per chain allowing invoke arbitrary opaque contracts
  • Exchange: the most used operation, allowing the provision of some tokens in order to receive other tokens in return

A detailed description of each instruction is described later, but the most important aspects are described in this section.

Spawn uses a path-dependant reserve transfer escrow/mint protocol to move funds.

Call is a way to represent a native call in encoding that is executable on each chain, and yet able to fill in (bind) some context details like the amount of assets and some addresses in a cross-chain (CVM) way.

Exchange requires configuration of the pool in CVM contract storage. So do mapping of native assets and chains’ routing capabilities available for CVM programs. Routing information describes the type of bridge and what capabilities and limitations connecting chains it has. There is no requirement for CVM contracts to be deployed on chains as long as chains are reachable by some form of execution.

Exchange/Transfer/Spawn are shortcuts for specific Calls. From here onwards, we will use Call to refer to any of these types of on-chain execution.

During program interpretation, while the instruction is executing, the result of call execution and bridging state (internal of Spawn) are recorded in executor state. In case of a CVM program error, funds are retained on the sending or receiving executor. Users can observe the state of the executor and send a program to handle the current state. For example, they can move funds to another account or chain.

Simplified CVM Design

/// CVM may be physically implemented differently on different chains
type OpaqueContract = SmartContract | Module
type OpaqueUser = OpaqueContract | Wallet

type Network = number

interface Program {
instructions: Instruction[]

/// random number to easy invocation tracing
type Tag = Uint8Array

/// allows one user to have several Executors to isolate funds and allowances
type Salt = Uint8Array

type Instruction = Transfer | Call | Spawn | Query | Exchange | Bond | Order | Abort | If

/// Is atomic without position recorded for user
/// Exchange - can be deposit into pool for LP token, Stake to get liquid stake token, borrow or lend.
/// So it is super set of what usually called swap over CFMM(AMMs).
/// Set `ExchangeError` to result register in case of fail.
interface Exchange {
in: (AssetAmount | BindedAmount)[]
min_out: AssetAmount[]
condition : ExchangeCondition?

/// in CFMM and OB, there are always fluctuations,
/// allows to decide if steel want to exchange if difference from
/// window on average is too far /// in CFMM and OB, there are always fluctuations,
/// allows to decide if steel want to exchange if difference from
/// window on average is too far
interface ExchangeCondition {
/// minimum number of blocks to consider averaging
window : number?
type : AggregationType

type AggregationType = TimeWeightedAverage

// time locks tokens, potentially getting some tokens out
// can be Stake or liquidity provision
interface Bond {
exchange : Exchange
/// maximum time for bonding, so as close as possible to it
time: Duration

/// intention for exchange which may happen in future
/// puts into ResultRegister either cross chain error error or OrderPositionId
interface Order {
exchange : Exchange
timeout : Duration
partial: Ratio[]

type OrderPositionId = Uint128

/// cross chain transaction/tracing identifier
interface ProgramInvocation {
/// for tracing
tag : Tag
/// network on which to execute program
network : Network

interface Spawn {
invocation: ProgramInvocation
program : Program
salt : Salt
assets : Assets

/// is target chain dependant
type Payload = Uint8Array

/// copies RegisterValue into program stack which drained into first Spawn
type Query = RegisterValue | Id

type RegisterValue = RegisterValue[]

type RegisterValue = ResultRegister | IPRegister | TipRegister | SelfRegister | VersionRegister | Carry

/// id sorted coints which was transferred into this `Spawn` (may be less then sent from original chain because if fee)
type Carry = AssetId[]

type IPRegister = Uint8

type TipRegister = Account

interface Call {
/// most chains have one default executor
/// tells how to interpret `payload`
executor : NativeExecutor?
payload : Payload
bindings : Bindings?

type NativeExecutor = Polkadot | Cosmos | Evm | CosmWasm | Solana

interface Binding {
index : Uint8
value : BindingValue

/// sorted keys by `index`
type Bindings = Binding[]

interface AssetAmount {
asset_id: AssetId
balance : Balance

/// amount from Carry register or part of it
type BindedAmount = {Carry , Ratio? }

type Account = Uint8Array

/// Self is account of Executor
type Self = Account

/// Pop - pops value from program stack into binding placeholder
type BindingValue = Self | Tip | Result | AssetAmount | GlobalId | Pop | Home

/// binds to chain id which originated message on this chain
/// can be used just to spawn back
type Home = NetworkId

/// Aborts transaction on execution chain if BindingValue contains Error.
/// In comparison with ResultRegister just set to error and stop executing (with funds retained on CVM Executor),
/// this instruction rollbacks changed made by this transaction (including results and funds) up to sender network if possible.
interface Abort {
abort_of_error: BindingValue

/// transfer from Executor account to
interface Transfer {
assets : Assets,
to: Tip | Account

/// sorted with unique ids
type Assets = [AssetId, Balance][]
type AssetId = GlobalId | LocalId
type GlobalId = Uint128
type LocalId = Uint8Array

type Balance = { AbsoluteAmount, Ratio}

type AbsoluteAmount = Uint128

type Unit = Uint128 Ratio
/// parts of whole
type Ratio = { numerator : Uint64, denominator: Uint64}

/// deterministic encoding of user wallet on each chain (virtual wallet)
interface UserOrigin = {
account : Account
network: Network

type ExecutorOrigin = Account

/// data send to other chain
interface SpawnPackage = {
Executor_origin : ExecutorOrigin
user_origin: UserOrigin
/// used by the while instantiating the Executor
/// so user can have several Executors on same chain or reuse existing (same salt)
salt: Salt
program : Program
assets: Assets
stack: ResultRegister[]

type ResultRegister = Error | ExecutionResult

type Error = CallError | TransferError | SpawnError | QueryError | OrderError | BondError | ExchangeError

/// open set of well know exchange errors
type ExchangeError = Uint8

type ExecutionResult = Ok | bytes
type Ok = '0'

/// One or more ordered set of unique cross chain accounts.
/// Owners have full control over funds in CVM Executor
type Owners = Identity[]

/// Network prefixed cross chain native chain Account
type Identity = [Network, Account]

/// this happens in Executor
function execute_program(caller: Account, program: Program) {
ensureOneOf(caller, this.owners, this.gateway)

// reset from the last execution
this.InstructionPointer = 0;
this.TIP = tip;

for (var instruction in program.instructions) {
this.InstructionPointer += 1;
this.result = this.execute_instruction(instruction);
if (is_error(this.result)) {

The model above does not map exactly what chains receive as bytes that are being input and then outputted to the bridge, rather it outlines the correct semantic model of execution.

Detailed CVM Design

Here are more details for instructions, roles, and the security model of the CVM.

Architecture and Flow

Here is the logical state of the CVM:

So, the user (wallet) has an executor (CVM account) on each chain (instantiated on demand).

The CVM account delegates a CVM contract to bridge assets and execute on behalf of the user across chains.

The CVM is bridge agnostic, with full capabilities when the target chain allows permissionless contracts and general message passing, as well as limited shortcuts when no custom code is allowed and only a subset of target cross-chain operations are possible.

Let’s look into one hop from one chain to another chain in detail (additional hops will just repeat part of the picture):

The diagram starts with a fully CVM-enabled flow.

In the end, it shortcuts VM execution to a standard protocol without doing SubProgram execution. This happens if there is an incentive to do a shortcut (for example, gas costs or limited CVM support on the target) and the subprogram is simple, like transfer (or another build in a cross-chain protocol).

All the CVM receives is CVM programs. It then uses routing configuration (of assets and bridges, and their features) to dispatch programs over underlying protocols.


Spawn instructions are executed concurrently. All other instructions are executed sequentially in a subprogram.


This executes a payload which is an ABI call to an opaque contract within the execution context of the chain.

The call instruction supports bindings values on the executing side of the program by specifying the Bindings. This allows us to construct a program that uses data only available on the executing side. Binding values are lazy filled in on the target chain.

The binding value Self injects the account in the Executor into a call.

Besides accessing the Self register, BindingValue allows for lazy lookups of AssetId conversions. This is done by using BindingValue::AssetId(GlobalId), or lazily converting Ratio to AbsoluteAmount type.

Bindings support byte aligned encodings (all that are prominent in crypto).


For example, the swap call of the following contract snippet expects a to address to receive the funds after a trade.

function swap(in_amount: number, in_asset: String, min_out: number, out_asset: String, to: AccountId) { 
// ...

If the caller wants to swap funds from the Executor account and receive the funds into the Executor account, they need to specify the BindingValue Self, using the index of the to field for the payload being passed to the contract.

The sender chain spawns with swap(100, "dot", "eth", 1, BindingValue::Self) call, and on target chain Self replaced with address on execution Executor.


Sends a Program to another chain to be executed asynchronously. It is only guaranteed to execute on the specified Network if its Program contains an instruction to execute on the Network of the Spawn context.

If Spawn fails, it may fail on the sender chain or on the receiver chain, but not both.

Funds are retained in the Executor.

ResultRegister is written.

AssetId in Assets are converted from the sender to receiver identifiers by the Executor.

How Spawn fails

Failure on the sender chain is asynchronous too.

There are two types of failures.

  1. When SpawnPackage fails fully without even calling the executor, in this case all funds are rolled back to the executor. Target chain will not have an on-chain trace of execution.

  2. When SpawnPackage reaches the target chain, starts execution in the Executor and fails. In this case ResultRegister is filled on the target chain and funds remain there. The sender chain considers Spawn to be a success.

There no in-between scenario when the failure is spread onto both chains.


Queries register values of a CVM instance across chains. It sets the current Result Register to QueryResult on the semantics of registers and RegisterValues.


The amount of assets can be specified using the Balance type. This allows foreign programs to specify sending a part of the total amount of funds using Ratio. Or, if the caller knows amount of the assets on the destination side using AbsoluteAmount.


Each executor keeps track of persistent states during and across executions, which are stored in different registers. Register values are always updated during execution and can be observed by other contracts.

Result Register

The result register contains the result of the last executed instruction.

If ResultRegister was set to Error and there is Restoration register containing a CVM program, it will be executed.

IP Register

The instruction pointer register contains the instruction pointer of the last executed program and is updated during program execution. Querying for the IP and Result can be used to compute the state of the Executor on another chain.

Tip Register

The Tip`` register contains the Account` of the account which incentives relayers pay gas fees for program propagation from chain to chain and execution. This can be the IBC relayer or incentive protocol.

Self Register

The self register contains the Account of the Executor. Most implementations will not need to use storage, but have access to special keywords, such as this in Solidity.

Version Register

The version register contains the semantic version of the contract code, which can be used to verify the subset of CVM functionality supported by the contract. Implementations that support upgradable contracts MUST update the version register. Functionality advertised through the version register MUST be supported by the contract.

Program Execution Semantics

Execution of a program is a two-stage process.

  1. First, the virtual machine MUST verify that the caller is allowed to execute programs for that specific instance, by verifying that the caller is one of the owners.
  2. Second, the instructions are iterated over and executed.

Implementers MUST execute each instruction in the provided order and MUST update the IP register after each instruction is executed. After each instruction is executed, the result register MUST be set to the return value of the instruction. The executor SHOULD NOT alter the return values but store them as returned. Because the return values are chain-specific, the actual structure is left undefined.

If an error is encountered by executing an instruction, the defined transactional behavior for that instruction should be abided by. All instructions defined in this documentation require the transaction to be aborted on failure. However, subsequent addendums may define new instructions with different behavior.

After the final instruction has been executed and registers are set, the execution stops and the transaction ends.

See Appendix A for the algorithm.

CVM Execution Semantics

Each chain within the CVM contains a singleton entity consisting of the Router, and the Gateway. Implementors MAY choose to create a monolithic smart contract or a set of modular contracts.


Each chain contains a singleton bridge aggregator, the Gateway, which abstracts over transports.


Each program arriving through the Gateway is passed to the Router, which becomes the initial beneficiary of the provided Assets before finding or instantiating an Executor instance. The router then transfers funds to the Executor instance.

Subsequent calls by the same Origin will not result in an instantiation, but instead in a re-use of the Executor instance. This allows foreign Origins to maintain state across different protocols, such as managing LP positions.

If no Executor instance has been created for a given caller, the call to the Router must either come from the IBC, XCM, or a local origin. After the instance has been created, it can be configured to accept other origins by the caller.


For a given CVM program, its Executor instance is derived from Network Account Salt. This allows users to create different Executor instances to execute programs against. Note that the Salt is not additive and only the composite Network Account is forwarded to remote chains as the user origin:

spawn network=A salt=0x01 { // the origin for the instructions is (A, AccountOnA, 0x1)
call 0x1337 // Call instruction executed on A
spawn network="B" salt=0x02 {} // Sub-program spawned on B, with 0x02 as salt, the origin for the instructions is (A, AccountOnA, 0x2)

A possible usage is to allow one program execution to act on the state of another program execution to restore funds.

In the CVM program above, the parent program salt 0x01 is not a prefix of the sub-program salt 0x02. The user is able to make its Executor origin using a fine grained mode. The following program is an example on how we can spread a salt:

spawn A 0x01 {             // Parent program spawned on A, with 0x01 as salt, the origin for the instructions is (A, AccountOnA, 0x01)
call 0x1337 // Call instruction executed on A
spawn B 0x0102 {}, // Sub-program spawned on B, with 0x0102 as salt, the origin for the instructions is (A, AccountOnA, 0x0102)

In next program, all spawned instances on all chains share state (including assets):

spawn A 0x01 {
call 0x1337
spawn B 0x01 {}, // Sub-program spawned on B, with 0x01 as salt, the origin for the instructions is (A, AccountOnA, 0x01) allows to share


Executor instances maintain a set of owners.

Programs are only executed by the Executor if the caller is in the set of owners.

On initial instantiation of the CVM Executor, the calling Identity is the owner. This can be a local or foreign account, depending on the origin. The owning Identity has total control of the Executor instance and the funds held can make delegate calls from the instance's account.

Oftentimes, multiple Identities represent a single real-world entity, such as a cross-chain protocol or a user. To accommodate for shared/global ownership of resources, each Executor keeps track of a set of Identities, which share ownership of the Executor. Each owning Identity has full permissions on the Executor instance.

Owners may be added by having the Executor call the appropriate setters. We will consider adding specialized instructions later. Owners may be removed by other owners. An CVM instance MUST always have at least one owner.


Execution fees are opt-in and paid by the user by using the Tip registry value.

The following example program performs an operation, and rewards the tip address:

"call" :
<Call> 0x13371337...
<Transfer> <Tip> { USDC: 15000000000000 }

Configuration and routing registry

Registry stores information about the mapping of native identifiers to CVM such as an asset and inter chain connectivity on chain.


The Spawn instruction uses the registry to lookup CVM cross chain capability with connection. Per chain registry stores its immediate neiborhood info, and may store more hops.

Propagating updates across registries is handled by the CVM too.

Assets and other CVM identifiers

Each asset identifier AssetId is 128 bit number in which the 32 initial bits represent the network identifier NetworkId. It is impossible for a scenario where the same asset id is different on another chain. To put it simply, it will not be possible that on one chain 123213 means PEPA and on other chains 123213 means SHIB. The Prefix allows you to find which network to look at for asset information as each chain introduces new assets independently.


Each chain contains data which maps assets to their local representations, such as ERC20 address. The Transfer instruction uses the registry to look up the correct identifiers.

Security Considerations

Ensuring that the caller is an owner is an incredibly important check, as the owner can delegate calls through the Executor, directly owning all state, funds, and possible (financial) positions associated with the Executor account. Since each Executor has their own Identity, they might own other accounts as well. Thus the owners control more accounts than just the contract storing the owners.

The Call instruction has the same security risks as calling any arbitrary smart contract, such as setting unlimited allowances.

Adding an owner to the set of owners grants them the ability to evict other owners.

Failure to execute an instruction will lead to a transaction being reverted, however, the funds will still be in the Executor account's control. Ensure that changing ownership is always done atomically (add and remove in the same transaction) to ensure funds are not lost forever.

Using bridges is equivalent to adding them as owners on your Executor instance.

In general different security can be applied to different programs.

Anonymous programs

These programs operate only on funds inside a program and a limited set of instructions can be executed without sender authentication.

A specific case of this occurence can be in a program that consists of Transfer, Spawn, Exchange only on assets transferred.

Cross protocol verification


When a program needs to transfer assets in the IBC, it will use the ICS20 protocol. In order to execute remote transactions on behalf of an account, it can use ICS27. In both packets, the same program can be sent as part of the batch and verified on the other end to be exactly the same when assembled for execution. For this, if one protocol is compromised we are still able to validate via the second one.

Trusted topology

A program can be executed if these are sent only from some subset of trusted channels.

Cross chain multisignatures

In this case, a program can be executed if it was sent by several chains.


For operations of high importance, the EDSCA signature of the program can be propagated from the sending chain and verified on the target chain.

Cross chain Transfer

Transfer is most common operation in blockchain, that why it deserver to be more detailed.

CVM on CW Cosmos to Cosmos uses ICS20 for value transfers. Each ICS20 channel must be upserted into CVM config.

CVM uses ICS-20 assets transfers on Cosmos chains. On Polkadot and Ethereum it uses escrow/mint, path dependant semantics compatible with ICS-20.

Data encoding

There is no single encoding used across all CVM hosts.

But encoding is always deterministic(like SCALE) or using deterministic subset of well known encoding(for example protobuf).

Specific encoding usage is subject to price, performance and usability constraints decided per chain to chain connection. Please look into code and/or indexers.


CVM is deployed in different hosts which are different in their capabilities.

CVM handles differences without chaining program semantics and reaches chains without contracts too.

Security of such multi deploy is described in other section.


Can be considered as 3 layers,

  1. Full deployment of contract with all capabilities. Can do anything.
  2. Partial fulfillment of CVM spec using on chain features in case not deployment is possible (usually can multi hop transfer, swap).
  3. Edges which may start or terminate CVM programs, but do not execute anything except simple sings (like one hop final transfer). Shortcuts capable doing only limited subset of operation on top of existing cross chain protocol.

For each chain and protocol it makes pragmatics solution to use existing liquidity and execution primitives.

No support for arbitrary contracts

Some chains do not support arbitrary contracts, but support a limited subset of instructions. In this case, only programs which use a limited subset of instruction will be executed on the target chain via virtual spawns.


The Cosmos Hub complies with the IBC ICS Atomic swap spec, but does not host a contract runtime.

In this case, programs trying to reach the Cosmos Hub from other chains will not spawn full programs on it. However, it will send only swaps and handle the invocation return on the sender chain.

No support for contract postconditions

Some chains cannot abort transaction based on arbitrary check after ABI invocation. In this case for specific subset of instructions to specific whitelisted contracts list will be allowed.


On Near, you cannot abort a Swap transaction if the amount is less than the expected limit. In this case, only trusted Swap contracts will be callable.


This documents shows some usages example of CVM.

Unwrap token

Shorten path of wrapped token.

Just spawn on unwrapping chain and spawn to target chain for execution.

Bond (stake, lock)

Call Bond instruction or do raw Call for Bonding.


Program to Bond on Stride and transfer staked token to Osmosis is detected as pattern expressed in CVM.

That part of program is translated to IBC calls to Stride without contracts deployed.

So this program is possible

execute amount=100atom { // on osmosis
spawn Osmosis amount=100atom {
spawn CosmosHub amount=All {
spawn Stride amount=All {
bond amount=All
spawn Osmosis amount=100%stAtom {
spawn Composable Cosmos amount=All {



A concrete example of using the CVM protocol is to transfer funds to a different chain, use them as collateral in a loan, transmit funds back to the source chain, and use them there. For this example, we'll omit querying for current account health and repayments.

Concretely, we want to execute the following operations:

  • Transfer funds to chain XYZ.
  • Call a smart contract to take out a loan.
  • Reward the relayer, to incentivize execution.
  • Send funds back.

Since we might not know the current interest rates, we'll use relative values for fund transfers, instead of absolute ones.

For this example, we have the source initiator be a regular user, however, a smart contract is capable of executing the same operations.

Although these operations are quite complicated to code by hand, using the CVM protocol, we can very succinctly express them:

spawn "XYZ" 0 {
call "0x1337" // chain-specific encoding to make a smart contract call
transfer tip=50usdc // 50 bucks for the fee. The relayer earns this if the inner spawn is dispatched
spawn HOME 0 amount=ALL { // We send over 100 DOT from ABC to XYZ.
transfer tip=50usdc // Another 50 bucks fee for the operation, but now reverse direction.
transfer to=Alice amount=All // On ABC, we transfer all USDC to the user.

Cross-Chain Exchange (swap)

Traditionally, users and applications were trading tokens that were only available on their native chain. If you were operating on a chain X, you would only be able to swap tokens that were registered on X (be it native, ERC20 virtualized or even virtualized and wrapped like WETH).

Manipulating tokens is very complex already. Users willing to move assets between chains are facing incredible difficulties. Not only bridging is hard, but also insecure. In most cases, the bridges are centralized and hackable by design. We, at Composable, try to push the blockchain vision forward, trustless from the start to the end.

In this example, we will execute a cross-chain swap through CVM and understand how programs are relayed, instructions executed and funds transferred. Be aware that this use case can be generalized to any DeFi protocol and it is the reason why CVM makes protocols cross-chain native.

Under CVM, tokens are free to fly between any chain connected to its network. Not only they can be traded regardless of their origin, but they are completely abstracted thanks to a globally unique CVM asset identifier.

Alice could submit the following CVM program, along 250 PICA to execute a cross-chain swap:

spawn network=Osmosis amount=250PICA { // Move to Osmosis with 250 PICA 
transfer to=Tip amount=25PICA // optional tip to the relayer for the journey
// Execute a swap, effectively trading 200 PICA for OSMO with 1% slippage tolerance.
// This might be replaced by an arbitrary `call 0x042502` representing the swap,
// but for some well-known protocols, we decided to include a custom, strongly typed instruction.
exchange give=200PICA want=99%OSMO
// At this point, we don't know how many OSMO/PICA we have.
// But we can ask CVM to move 100% of both!
spawn network=Picasso amount=100%OSMO,100%PICA {
// optional tip the relayer for the cosy home with the remaining PICA.
transfer to=Tip amount=100%PICA
// 1.3.2. Funds are safu.
transfer to=Alice amount=100%OSMO
  1. Alice submits the CVM program and the instruction 1. is executed, resulting in:
    1. the child CVM program, consisting of the instructions [1.1., 1.2., 1.3.] is being submitted within an IBC packet to Osmosis.
    2. the funds attached to the child program, 250 PICA, are being transferred to Osmosis using an ICS20 transfer.
  2. An IBC relayer, listening to on-chain events, determine that relaying the IBC packet containing the CVM program is profitable. It proceeds and relays both the funds and the packet to Osmosis.
  3. The packet is being submitted by the relayer and subsequently processed on Osmosis, resulting in the child CVM program being executed:
    1. the instruction 1.1. is executed: 25 PICA are transferred to the relayer.
    2. the instruction 1.2. is executed: 200 PICA are traded against an unknown amount X of OSMO.
    3. the instruction 1.3. is executed, resulting in:
      1. the second child CVM program, consisting of the instructions [1.3.1., 1.3.2.] is being submitted within an IBC packet to Picasso.
      2. the funds attached to the second child program, 100% of the OSMO and 100% of the remaining PICA are being transferred to Picasso using an ICS20 transfer.
  4. Finally, an IBC relayer determines that relaying the program is again profitable, the packet and the funds are relayed to Picasso.
  5. The packet is being submitted by the relayer and subsequently processed on Picasso, resulting in the second child CVM program being executed:
    1. the instruction 1.3.1. is executed: 100% of the remaining PICA (25 + dust from the swap) is transferred to the relayer.
    2. the instruction 1.3.2. is executed: 100% of the OSMO is transferred back to Alice.