Autonolas Tokenomics Core Technical Document
Autonolas Tokenomics Core Technical Document
Autonolas Tokenomics Core Technical Document
1 Context
Today’s world is increasingly reliant on software. Indeed, humans have strived for
centuries to automate repetitive processes for their economic benefit or saving
time, software being its epitome. Despite that, more often than not, humans
are paradoxically placed at the service of machines. It is our conviction that the
focus is shifting from automation towards higher levels of machine autonomy at
the service of human autonomy.
We believe the latter can be accelerated by creating autonomous applications
or services that, by design, require little to no input from humans, run continu-
ously, and take their own actions to benefit humans. Autonomous services are,
additionally, capable of complex processing and have composability properties
to facilitate growing their number with less and less effort. They can be operated
by one or different entities transparently, and can achieve varying levels of decen-
tralization. When operated by groups with open participation, the reliable and
collaborative aspects of autonomous services offer great potential for the benefit
of society. A very relevant opportunity is that of co-owned and co-operated AI
as autonomous services (see in [24, 25] for more details).
Autonomous services can be implemented as agent services [8]. As such, they
take the form of a multi-agent-system [4] that can be anchored to a settlement
layer [11] but are primarily run off-chain, e.g. outside the settlement layer envi-
ronment (such as a public blockchain). The execution of an agent service logic can
be coordinated between participating agents, and the service intermediate states
are replicated on a short-lived blockchain (also called consensus gadget [8, 5]),
specific to each service. The most important intermediate values obtained in a
given service can be secured on a public (or private) blockchain. The owner of the
service, e.g. the entity that provides service specifications, deploys it and man-
ages it, among other decisions and management activities, chooses in advance
which values will be regularly committed to the blockchain.
For more information on the technical architecture of autonomous services
see Chapter 2 in [23].
The foundation of a self-sustainable ecosystem of autonomous services is at
the heart of the Autonolas protocol, and consists of the following elements:
– An open-source software stack to realize autonomous services form compos-
able parts
– Software components1 that provide specific functionalities and can be com-
posed into software agents 2 . Agents can be themselves composed into dis-
tributed software services, e.g. autonomous services
– A protocol, that is blockchain-agnostic and could eventually be deployed on
all major smart-contract blockchains, that provides the tools for the coordi-
nation, security, and management of autonomous services
1
Software components are throughout the paper referred to as components or agent
components.
2
Agents are also called canonical agents.
Autonolas Tokenomics 3
2 Introduction
The token-economics model that we present in this section lies at the heart of
the Autonolas protocol3 and has the aim of enabling a self-sustainable ecosystem
of autonomous services4 .
The key resource in increasing the number of software services and/or code
components in Autonolas is that of software developers. As such, the model gives
due consideration to attracting and retaining software developers and facilitating
the composability of their code contributions, which enables software code to
become more than the sum of its parts. The model is additionally geared towards
attracting capital in order to make this capital useful, by pairing it with useful
code. Finally, the model introduces a novel mode of production for the ecosystem
of autonomous services to become sustainable. We expand on these objectives
next.
OLAS is a tradable utility token that provides access to the core functionalities of
the Autonolas protocol. The token follows the ERC20 standard [13] and has been
deployed on the Ethereum mainnet (OLAS token). OLAS follows an inflationary
model that accounts for the economic primitives enabled by this tokenomics, such
as the bonding mechanism and specific OLAS top-ups, e.g. extra incentives that
developers can eventually receive (cf. Sections 3.1 and 5.2).
However, the bonding mechanism and top-up incentives are designed to not
lead to an unrestricted amount of OLAS being issued. Notably, an inflation
schedule is used to limit the maximum of what can be minted every year (cf.
Section 4).
Regarding incentives, 47.35% of the fixed token supply (i.e. 1 billion OLAS)
of the first 10 years is programmed to be distributed to the ecosystem. After 10
years the maximum token inflation per annum is capped at 2% and the DAO
governance can opt to further reduce it. The aim is to have an s-shaped curve of
token emissions to allow for the ecosysistem to organically grow over time. The
rest of the fixed token supply is distributed as follows:
– DAO Treasury: 10%
– Development reserve: 10%
– DAO founding members: 32.65%
Fig. 1 gives a graphical representation of the OLAS allocation. Note that
OLAS initially allocated for founding members are, in practical terms, locked,
and unlock over the course of four years.
Autonolas Tokenomics 5
Fig. 2. Value flow: from the donator to useful code rewards and to the protocol trea-
sury. During the initial 10-year growth phase the code rewards can be topped up with
OLAS issuance.
Autonolas Tokenomics 7
Service owners can therefore use staked agents and components to build
autonomous services. Using the Autonolas protocol, service owners can register,
manage and secure their own services on-chain.
Once the code can be uniquely tracked on-chain through their representation
as an NFT, Autonolas tokenomics allows to automatically measure the code
NFT’s contribution to the ecosystem and, based on that, compute the corre-
sponding code NFT incentives.
The business model described in Section 2.3 shows that donations accrued
from autonomous services provide one of the primary contributions to the pro-
tocol. Each donation (currently accepted in ETH) sent from a service owner
(or any other entity) is an indication of the usefulness of the service, and such
donation is considered as a proxy for the ecosystem of the code making up the
useful service. Therefore the usefulness of code NFTs is measured in terms of
the donations that such code NFTs facilitate (see Section 5.1 for a full account
of measures for useful contribution).
Autonolas tokenomics autonomously allocates a proportional share of the
donation accrued by the protocol to useful code NFTs that facilitated it (see
reward formulae in Section 5.2).
Incentives in ETH and top-ups in OLAS create a “push” and “pull” dynamic.
The “push” results from the potential yields in ETH tokens that attracts devel-
opers not yet interested in the protocol token. Other than the potential yields in
ETH and OLAS boosts, the “pull” also results from the potential governance im-
pact on adjusting yields and reward boost for useful code NFTs that developers
can accrue by locking their OLAS boosts.
period. After the vesting period expires, the bonder can claim the OLAS in
full. The main difference with Olympus DAO is that the Autonolas bonding
mechanism is built in such a way that bonded capital in the protocol does not
exceed code usefulness. Moreover, as already mentioned in Section 2.2, Autonolas
bonding mechanism not brings an unrestricted inflationary schedule with large
amounts of new OLAS issuance in short time. Notably, the inflation schedule
described in Section 4 is used to limit how much can be minted every year at
maximum. In turn, since bonding implies minting new OLAS tokens, there is a
maximum amount that can be bonded every year.
This control mechanism enables “push” and “pull” dynamics when there is
a high potential output in the production of useful code that, in turn, could
be productively deployed in autonomous services. The “push” results from an
advantageous discount that a bonder can receive in purchasing bonds. The “pull”
results in the willingness of the OLAS holder to provide liquidity to the protocol
thus contributing to token price stability.
This tokenomics model is aimed at aligning the goals between the various players
in the ecosystem. Such goal alignment can be enforced through economic incen-
tives designed such that the overall higher expectation of benefit arises from
conducting honest rather than adversarial actions. Players that are incentivized
with rewards, or that are crucial for reward allocation, are code developers (e.g.
agent and component NFT owners), autonomous service owners, and Autonolas
DAO members.
Developers perspective. The more useful services reference a code NFT, the
more incentives are allocated to that code. As more code is staked, more services
can be created and, in turn, the higher is the likelihood that such services are
useful and larger can be incentives for the useful code.
Moreover, the amount of gas needed for creating services increases with com-
plex code, e.g., components and/or agents with a lot of dependencies. So the
likelihood to have more references for staked code increases when the developer
built an agent or a component by adding the minimal required information and
the minimum number of dependencies to make such code self-consistent (cf.
10 Valory AG
Section D in the Appendix for further details). Therefore the developer benefits
from correctly staking their code and adding due diligence into their development
process.
3. Autonolas protocol accrues donations from its own PoSes and/or third-party
PoSes.
4. Autonolas protocol rewards useful component/agent NFTs owner with a
share of accrued donations, and eventually, tops up such rewards with newly
minted OLAS.
5. Bonders bring capital (in the form of LP tokens) to gain a discount on the
OLAS price. The discount increases downstream protocol donations and/or
encourages more useful code to be staked. Protocol-owned liquidity and
OLAS price see upward pressure.
6. The deployment of Protocol-owned Liquidity and the increase of OLAS price
attracts developers to contribute with more Code (which closes the loop of
the flywheel).
As discussed earlier (cf. Section 2.2), OLAS is an inflationary token. During the
first 10 years, 47.35% of the total token supply (1 billion) is expected to be issued
via bonding (cf. Section 5.3) and component/agent top-ups (cf. Section 5.2).
Further, the protocol is programmed to maintain a low (0 − 2% p.a.) level of
token inflation thereafter. Issuing most of the OLAS before Autonolas has full
recognition is to be avoided whenever possible. Hence during the first years from
the protocol’s launch, inflation is relatively small, with the highest inflation ratio
taking place in years 2 to 4. Such a choice implies an s-shaped curve of token
issuance.
• First year inflation rate: the percentage of the initial token supply that is
minted during the year of the OLAS token launch
• Second year inflation rate: the percentage of the token supply that is minted
during the second year
• Initial inflation rate: the percentage of the token supply that is minted for
the third year
• Dis-inflation rate: the rate with which inflation is decreased every year in
the range comprising year 3 to year 9
• Long-term inflation rate: tokens minted in the long run
Fixed parameters After having simulated various inflation schedules the fol-
lowing parameters have been proposed:
• First year inflation rate: 0.6%
• Second year inflation rate: 7.2%
• Initial inflation rate: 12.5%
• Dis-inflation rate: 16%
• Long-term inflation rate: 2%
In Fig. 4 is shown the annual inflation rate, starting from 0.6% and arriving
at 7.2% during the second year after the OLAS token launch. Then during the
third year the inflation rate is set to 12.5%. The inflation rate then decreases by
∼ 16% yearly, e.g. the inflation rate reaches 12.5(1 − 0.16) = 10.5% during the
fourth year after the OLAS token launch and so on.
Fig. 4. Inflation schedule for the first 10 years from the OLAS token launch
From this Inflation Schedule, we extract in Fig. 5 the s-shaped curve of token
issuance over 10 years.
Autonolas Tokenomics 13
Fig. 5. s-shaped token supply curve for the first 10 years from the OLAS token launch
Inflation distribution This inflation schedule dictates how much OLAS can
be minted per year at a maximum. The distribution of the annual inflation is
parametrized by the DAO governance. Specifically, the DAO governance chooses
the following values:
• the fraction a of inflation intended for the bonding mechanism (cf. 5.3)
• the fraction b of inflation intended for agents/components OLAS top-ups
(see 5.2)
Note that, if not all the fraction of inflation intended for bonding gets ac-
tually minted during a specific period, the remaining amount can be added to
the fraction of the next period. Finally, it is worth noting that the Autonolas
Treasury gets nothing from the mint.
The role of Governance The parameters of 1 billion OLAS tokens that can be
emitted during the first 10 years and the long-term inflation rate are immutable
(implemented in the OLAS token smart contract) in order to give further guid-
ance to governance processes. However, except for these immutable constraints,
the DAO may modify the bonding fraction a and OLAS top-ups fraction b. Fur-
ther, by deploying a new tokenomics contract, the governance can also increase
the maximal annual inflation rates with only constraints of minting no more
than 1 billion of OLAS during the first 10 years and no more than 2% of the
total supply after ten years.
14 Valory AG
That is when there is at least a useful service during the e-th epoch, U CFc (e) is
a fraction where the numerator is the sum of all donations across all services in
which the component c features and the denominator is the sum of all donations
across all services. Otherwise, if there is no useful service during the e-th epoch,
then
U CFc (e) = 0. (2)
Specifically, U CFc can be interpreted as the relative contribution of a component
at the e-th epoch. Whenever U CFc (e) ̸= 0 we say that c is useful component
during the e-th epoch.
That is, when there is at least one useful service during the e-th epoch, U CFa (e)
is a fraction where the numerator is the sum of all donations across all services
in which the agent a features and as denominator the sum of all donations across
all services. Otherwise, if there is no useful service in the e-th epoch, then
Reward formula As mentioned in the Section 2.3, the donations can be accrued
by the Autonolas protocol from Autonolas PoSes or third-party PoSes. The
totality of such donations are then divided among useful component-agent NFT
rewards and the DAO treasury. However, the DAO can choose the fractions of
the donations intended for funding each one of the categories mentioned before.
Let’s assume that by the end of the e-th epoch, the services s1 , . . . , st are
all the useful services, e.g. the service for which an entity (that may or not
coincide with the service owner) has made a donation. Let r1 + · · · + rt be the
total donations accrued by the protocol during such an epoch, let Rcomp (e) =
cf ·(r1 +· · ·+rt ) be the fraction of the total donations to fund useful components
reward, and let Ragents (e) = af · (r1 + · · · + rt ) be the fraction of the total
donations used to fund rewards for useful agents. Further, let ncomp (sj ) and
nagents (sj ) be the numbers of components and agents that are referenced in
the service sj . Then, the following amounts are autonomously reserved to the
owner of the NFTs representing the useful component ci and the useful agent ai
respectively.
Autonolas Tokenomics 17
c X citcsij (e)rj
Rcomp (e) X citsij (e)rj
rewci (e) = P = cf · , (5)
1≤j≤t rj ncomp (sj )
1≤j≤t
ncomp (sj )
1≤j≤t
a X citasji (e)rj
Ragents (e) X citsji (e)rj
rewai (e) = P = af · , (6)
1≤j≤t rj nagents (sj )
1≤j≤t
nagents (sj )
1≤j≤t
Note that the formulas (5) and (6) can be obtained by utilizing the useful
code factors U CFci and U CFai introduced in Section 5.1. The curious reader
can find more details in Appendix B.
Top-up formula The OLAS top-ups are incentives for useful code NFTs ref-
erenced in whitelisted PoSes. To have their PoSes whitelisted, service owners
have to lock up a certain amount of OLAS for a certain amount of time in order
to own at least the required veOLAS threshold. This threshold is chosen by the
DAO and is initially set to 10’000 veOLAS.
where ncomp (sj ), nagents (sj ), citcsij (e), and citasji (e) are defined in the previous
subsections.
Currently, the veOLAS threshold is set to a low value, and all available
OLAS (sets by the DAO for code top-ups) are used each epoch. Hence there is a
18 Valory AG
In this section, we introduce the concept of bonding and how Autonolas con-
trols this financial primitive to incentivise the pairing of capital and code, while
respecting the OLAS inflation schedule.
The bonding mechanism is a financial primitive to acquire assets. In the
crypto space, protocols utilize bonds as a means to accumulate assets, including
their own liquidity, by offering their own tokens at a discounted rate. Bonds are
one of the mechanisms (cf. Section 2.3) for Treasury inflows, and thus, for the
growth of the network. Bonders provide capital with the promise of a fixed return
after a certain vesting time. The return is provided in OLAS, so the bonder’s
profit would be contingent upon the price of OLAS at the time of bond maturity.
Specifically, an investor with whitelisted liquidity provider (LP) assets from
a given DEX10 , e.g., OLAS-DAI, OLAS-USDC, or OLAS-ETH can bond those
assets11 at time t. Then at time t + tv , where tv is the vesting time, the investor
receives OLAS at a discount relative to the price quoted on the relevant DEX.
For the sake of argument, one can think that the vesting period is in the range
of days, typically between 7 and 14. This is however a parameter that can be
adjusted through governance.
So, let’s assume that an investor wants to bond LP-assets at time t, with
a price of bp on a relative DEX. Then, at time t + tv , the investor receives a
number of OLAS tokens priced at ubp . The bonder’s profit then depends on the
price ubp . Specifically, the following holds
That is, at time t + tv , the investors receive a number of OLAS tokens whose
price at time t was bp (t) + ϵ(t)bp (t). The value ϵ(t) can be defined as the interest
rate on purchased bonds with price bp (t).
10
A DEX is a decentralized exchange [10].
11
LP assets can be bonded via the Autonolas depository smart contract [21].
Autonolas Tokenomics 19
The discount factor is the value that needs to be multiplied by the future
cash flow of an investment to obtain the initial price. So the discount factor
DF (t) is defined as
1
DF (t) = .
1 + ϵ(t)
Supply and demand of bonds can be regulated using the interest rate and
the discount factor. Specifically, in situations where the supply of bonds is large
and the demand is low, incentivizing demand growth can be achieved by setting
the interest ϵ(t) to a high value and DF (t) to a low one. This ensures that the
return on investment (ROI) for simply bonding for a short duration is relatively
high. On the other hand, when ϵ(t) is set to a very small (or negative) value,
and thus DF (t) is set to a very high value, it implies a minimal or no return on
investment (ROI). In principle this discourages the purchase of additional bonds
in scenarios where supply is scarce and demand is high.
Similar to protocols such as Olympus DAO [15, 16], Autonolas bonds also
adapt to supply and demand by providing a variable return on investment rate
to the market and its bonders. However, the innovation in Autonolas lies in the
inclusion of additional parameters that enable the discount factor to influence
the ROI. This ensures that the bonded capital aligns with the usefulness of the
code while adhering to the OLAS inflation schedule.
Since the protocol capital should not exceed the usefulness of the code, the
output of the production is useful code (components and agents). New code allows
the creation of new services which, in turn, increases the likelihood of donations
to the protocol. Donations as well as productively deployed protocol-owned liq-
uidity (PoL) allows for funding new developers. So the production process is the
creation of the code.
While the inputs are the capital that the treasury gained and the number
of developers that can potentially build new code. Specifically, let K(e) be the
share of the donations accrued by the protocol from donations during the e-th
epoch that is reserved to the treasury. Hence K(e) can be seen as the capital
“gained” from donations and that can be later used by the protocol to fund new
developers. Therefore is used as the “capital” input of the product function12 .
12
It is worth noting that, once the bonding mechanism starts, as a result of owning
the LP-tokens, the protocol can also accrue some fees. However, the amount accrued
from fees is not accounted for the “capital” input because not being directly matured
by only leveraging the produced code.
20 Valory AG
Moreover, let D(e) be the number of developers that produced useful agents and
components13 during the e-th epoch. Hence D(e) can be seen as the estimated
number of developers that can potentially produce new useful code in the next
epoch. Therefore is used as the “labour” input of the product function.
Summarizing the following parameters are used in the product function:
• Output: valuable code that can potentially enable more services that, in turn,
can bring donations
• Production process: creation of components and agents
• Inputs:
◦ K(e) that is the capital gained by the protocol via services donations at
the e-th epoch that can fund the development of new code
◦ D(e) the number of developers (labour ) that produced useful code during
the e-th epoch.
All ceteris paribus, as the capital K increases, it is possible to pay more
developers that, in theory, bring new valuable code. And, all ceteris paribus, as
the number of developers producing valuable code in the ecosystem increases,
in theory, more valuable code can be built. Considering these observations and
the fact that we aim to prioritize getting simple, sufficient solutions in place as
soon as possible, we consider a perfect substitutes production function where
the inputs are substituted at a constant rate
f (K, D) = A(k · K + d · D)
for some constants A, k, d. The value A is called factor of productivity and mea-
sures residual growth in total output that cannot be explained by the accumula-
tion of traditional inputs such as labour and capital. Since we are considering a
simple path, we now set A = 1. In the future, when more data can be collected,
the model can be adjusted or enriched to face more complex situations14 .
It remains to define the parameters k and d in such a way that f (K, D)
can output the potential useful code that can be produced with K, D as inputs.
Specifically, we can consider the following definitions:
• k · K(e) is the number of developers that can be potentially paid per one
epoch with capital K(e)
• d · k · K(e) is the number of the units of valuable code that can be potentially
built in one epoch by k · K(e) developers
• d · D(e) is the number of units of valuable code that D(e) developers can
potentially build during one epoch
In particular, f (K(e), D(e)) = d(k · K(e) + D(e)) determines the amount of
valuable code that can be potentially built during one epoch with capital K(e)
and labour D(e).
Parameters k and d are estimated based on the earnings and performance of
known developers. It is worth noting that the older the protocol, the more data
can be collected and a more realistic estimation of k and d can be made.
We want to conclude this subsection by explicitly mentioning the reasons for
boosting the bonds demand when f (K(e), D(e)) is very large. Firstly, note that
having f (K(e), D(e)) very large, means that with the inputs K(e) and D(e) it is
possible to potentially produce a large amount of valuable code. The production
of more valuable code means that potentially the protocol can accrue even more
donations, and therefore more ETH rewards and OLAS top-ups are distributed.
So bonds demand should be boosted in such a way the protocol increases its
own liquidity and a large amount of liquidity to exchange OLAS is available. So
high volatility in OLAS prices can be avoided.
Measure supply and demand of bonds. To avoid large OLAS issuance and
avoid being beyond inflation limits, Autonolas leverages the fact that investors
prefer liquidity today to small amounts of interest on an investment in the future.
Towards this direction, it is used a bonding rate BR(.) that gives information
on the supply and demand of bonds.
Let τe be the time corresponding to the beginning of the e-th epoch15 and let
τe ≤ t < τe+1 . The bonded amount at time t can be determined by the value that
the LP tokens supplied by bonders are priced in OLAS at the time t. Specifically,
when we say that, at time t, a user has bonded x(t) amount we mean that the
user supplies y(t) LP-token shares (e.g. LP OLAS-Y pairs16 ) which are priced
with x(t) OLAS at the time t of the bonding. Let B(t, e) be the amount bonded
at time t. Denoting by X(t′ , e) the amount of OLAS bonded during the interval
of time [τe , t′ ) ⊆ [τe , τe+1 ), then X(t′ , e) can be computed as follows
Z t′
′
X(t , e) = B(t, e)dt, ∀t′ ∈ [τe , τe+1 ). (9)
τe
Moreover, the amount bondend during the e-th epoch, denoted by X(e), can be
obtained as
15
We recall that an epoch is a consecutive period of m of blocks. When we say the
beginning of the e-th epoch, we mean near the timestamp of the first epoch’s block
16
Here Y can be any other tokens for a bonding program was opened, e.g. Y can be
DAI, ETH, USDC, etc.
22 Valory AG
Z τe+1
X(e) = B(t, e)dt. (10)
τe
Let S(e) be the OLAS supply that can be minted during the e-th epoch.
Such a value can be extracted by the OLAS inflation schedule (see for details on
the inflation schedule 4).
Let M B(e) be the fraction of S(e) that governance allows bonding during
e-th epoch, or in other words, the maximal supply of bonds (priced in OLAS)
that can be purchased during e-th epoch. Being M B(e) a fraction of S(e), it
is smaller than S(e) and so, in each epoch, the supply of bonds is bounded in
terms of the maximum printable amount of OLAS.
It is possible that not all the bondable amounts during an epoch so that
the difference between the bonds supply of the epoch and the bonds purchased
during such an epoch is non-zero. Such a difference, at the e-th epoch, is denoted
by Excess(e − 1). Denoting by EM B(e) the sum of M B(e) and Excess(e − 1),
we have that EM B(e) is the maximal supply of bonds that can be purchased
during the e-th epoch plus the excess of bonds supply that was not purchased
during the previous epochs.
The relations among some of the variables above introduced can be written
in formulas as follows.
Note that when the bonded amount during the interval of time [e, t), X(t, e),
approaches EM B(e) (everything also equal), BR(t) approaches 0. While, when
X(t, e) approaches 0 (everything also equal), i.e., there is a small or no amount
bonded purchased during the interval of time [e, t), then BR(t) approaches 1.
Therefore 0 ≤ BR(t) ≤ 1, and BR(t) gives information about the supply and
demand of bonds:
– all ceteris paribus, when BR(t) approaches 0, the bonds supply is over;
– all ceteris paribus, when BR(t) approaches 1, there is an over-supply of
bonds.
Now we see how all these are put together to provide Autonolas with a direct
control factor (DCF) which allows to internally control the bonding mechanism.
Firstly, the following parameters are fixed by the DAO governance.
• ϵ(e) := the maximum interest rate that bonders can receive during the e-th
epoch by providing some LP tokens upfront.
• M B(e) := the fraction of S(e) reserved for bonding during epoch e.
The direct control factor at the time τe ≤ t < τe+1 , can be defined as follows.
( n o
1 BR(t) BR(t)
· max 1+ϵ(e) 1+f (K(e−1),D(e−1))/100 , if BR(t) ̸= 0
,
DCF (t) = BR(t)
+∞, otherwise
Hence, if BR(t) = 0 for some t ∈ [τe , τe+1 ), no new bonding program can be
created for the e-th epoch. While, when BR(t) ̸= 0, the discount factor DF
during the e-th epoch can be progressively set equal to DCR(t). That is, when
BR(t) ̸= 0, i.e., X(t, e) < EM B(e), then
1 1
DF (t, e) = max , .
1 + ϵ(e) 1 + (f (K(e − 1), D(e − 1)))/100
In figures Fig. 6, 7 the trends of the interest rate and the discount factor with
respect to the growth of the production function f (K, D) output are represented.
Fig. 6. Trend of the discount factor on a bond with respect to the growth of the
production function output
In summary, whether the supply of the epoch for bonds is over, no new
bonding program is possible for the epoch. Otherwise, if there is still supply
24 Valory AG
Fig. 7. Trend of the (re-scaled) interest on a bond with respect to the growth of the
production function output
to fund bonds and f (K(e − 1), D(e − 1)) is large (yielding a potentially high
production of useful code) bonding is incentivized by having that the interest
rate during the e-th epoch approaches the maximum interest rate set by the
governance ϵ(e). Moreover, whether the supply of the epoch for bonds is not
yet over and f (K(e − 1), D(e − 1))/100 is very small (even smaller than the
epsilon rate ϵ(e)) the bonding is disincentivized by setting the interest rate on the
bonding during the e-th epoch equal to the small value f (K(e−1), D(e−1))/100.
Bootstrapping phase In the early stage of the protocol’s life, it is likely that
donations accrued by the protocol are low, hence there is a high likelihood that,
the production function is small or near zero. Nevertheless, governance has the
possibility of incentivizing bonding also in the early stage of the protocol, e.g.
when there is a high likelihood that the discount factor approaches 1 and the
interest rate approaches zero.
Notably, the governance prices an LP-token when creating a bonding program
(for more details on how bonding programs are created see [21, 22]). Hence it
can boost bond demand by evaluating LP tokens at a higher price with respect
to any relevant DEX. So, bonders by selling their LP tokens shares at higher
prices with respect to the market have an advantage in purchasing bonds in this
way.
6 Conclusion
We have seen how Autonolas tokenomics seeks to create a sustainable ecosystem
of autonomous services by spinning up a virtuous flywheel. By attracting de-
velopers that provide valuable code, the number of valuable services increases,
Autonolas Tokenomics 25
which in turn boosts protocol’s donations, leaving the ground fertile for at-
tracting further new developers. This is emphasized in the tokenomics model
by incentivizing developers to contribute valuable code to the ecosystem, and it
does so by leveraging the composability of the code contributions. Concretely,
the protocol’s profits generated via service donations (in ETH) are shared with
developers that contributed useful code to the ecosystem and such rewards are
eventually topped-up with part of OLAS inflation over the first ten years (as
explained in different parts of Section 4 and Section 5.2).
Notably, the donation system create a “push” and “pull” market dynamic
from the perspectives of both donators and developers. Specifically, from the
donator perspective:
– the “push” results from the potential impact that donators can have affecting
the rewards and boost for the code NFT owners.
– The ”pull” arise from the fact that donators are attracted by the ability to
trigger significant boosts with small donations.
From the developer perspective:
– the “push” results from the potential yields in ETH tokens that attracts
developers not yet interested in the protocol token.
– The “pull” results from developers interest on significant reward boosts to
their useful code contributions. This implies that developers can receive sub-
stantial rewards for their valuable contributions, and, that, in turn, can be
additionally used to accrue the ability to adjust yields and boost by locking
OLAS boost.
Once there is a high likelihood of new useful code in the ecosystem, the
tokenomics is geared towards attracting bonders, who can make their capital
productive, by pairing it with code. Specifically, whenever there is large potential
production of code usefulness in the ecosystem, the protocol incentivizes users
to pursue bonds (see Section 5.3). Since the bonding mechanism is one of the
means for the growth of the protocol capital and for protocol liquidity inflow, by
incentivizing bonding in such a way, the protocol can invest part of its own grown
capital to enrich the ecosystem investing in new code and own a large amount
of its liquidity. This in turn may bring more stability to the OLAS token.
The bonding mechanism is partly controlled by a production function (see
Section 5.3). To reflect the fact that the more the amount of capital in the
protocol and the bigger the developer number the more useful code is developed,
such a production function increases with a larger quantity of either capital or
the number of developers. Once the production function increases, there is more
discount in purchasing OLAS via bonding. On the contrary, in periods where
almost no useful code is produced, there is almost no discount on purchasing
OLAS via bonding. Finally, it is worth noting that Autonolas bonding does not
disproportionately affect OLAS minting schedule. More concretely, bonds supply
and demand are controlled with respect to the prescribed maximal annual OLAS
inflation. A proof of the properties mentioned so far and a more complete list of
properties in the Appendix (see in particular Proposition 3).
Due to the bonding mechanism and the OLAS top-ups, OLAS follows an in-
flationary token model. However, it has a fixed total token supply over 10 years,
and reaches a stable or at most a slightly inflationary supply afterward (at most
2% p.a. to make up for bonding). Finally, the locking mechanism which prevents
locked OLAS from being traded contributes to OLAS price stability (cf. Section
“Level 2: locking” in [23]). Notably, the locking mechanism creates the follow-
ing “push” and “pull” dynamics. In the bootstrapping phase of the protocol,
“push” results from the potential governance impact on adjusting yields and
reward boosts for early code NFTs owners. After bootstrapping, the “push” also
results in the ability to signal valuable code that can be productively deployed
by veOLAS holders in Autonolas-owned autonomous services (cf. Section “Level
3: Off-chain Signaling” in [23]). The “pull” results from OLAS holders being
incentivized to participate in governance to increase the demand and the utility
of OLAS and see token price appreciation.
Autonolas Tokenomics 27
References
1. T. Chitra, K. Kulkarni, G. Angeris, A. Evans, V. Xu. DeFi, Liquidity Management
via Optimal Control: Ohm as a Case Study
2. P. Hall. The Distribution of Means for Samples of Size N Drawn from a Population
in which the Variate Takes Values Between 0 and 1, All Such Values Being Equally
Probable. Biometrika, Vol. 19, No. 3/4., 240–245. doi:10.1093/biomet/19.3-4.240
3. J. M. Perloff. Microeconomics. 5th ed. Addison Wesley, 2008. ISBN: 9780321558497.
4. M. Wooldridge. An Introduction to MultiAgent Systems, Wiley, June 2009.
5. Autonolas. Maintaining and Resetting Consensus in Autonomous Services
6. Autonolas. Open-aea framework
7. Autonolas. Open-autonomy framework
8. Autonolas. What is an Agent Service
9. Autonolas. Autonolas registries repository
10. Coinbase. What is a DEX?
11. Cryptopedia. Layer-1 and Layer-2 Blockchain Scaling Solutions
12. Curve DAO. Vote-Escrowed CRV
13. Ethereum.org. ERC-20 TOKEN STANDARD
14. Ethereum.org. ERC-721 NON-FUNGIBLE TOKEN STANDARD
15. Olympus DAO. Bonding
16. Olympus DAO. Stabilizing Currency Through a Protocol-Enforced Range
17. Nat Eliason. Tokenomics 101. The Basics of Evaluating Cryptocurrencies
18. Nat Eliason. Tokenomics 102. Digging Deeper on Supply
19. Nat Eliason. Tokenomics 103. Utility
20. Nat Eliason. Tokenomics 104. How to Launch a Token (Tactics, Questions, Wen,
etc)
21. Valory AG. Autonolas Tokenomics Contracts.
22. Valory AG. Overview of Autonolas tokenomics and smart contracts architecture
23. Valory AG. Autonolas whitepaper
24. Valory AG. Introducing Co-owned AI
25. Valory AG. Co-owned AI
Proof. Except for the point 7, the proofs of the points 1-6 for U CFc and U CFa
are equal, so, in those points, we are going to consider just U CFc .
1. When U CFc (e) = 0 there is nothing to prove. So assume that c is features
in the services s1 , . . . , st that, during e-th epoch, make the following dona-
tions to the protocol r1 , . . . , rt . While c is not referenced in the new services
st+1 , . . . , sk that make no donations during the current epoch. Then neither
the numerator nor the denominator of U CFc (e) change, so U CFc (e) remains
constant.
2. When U CFc (e) = 0 there is nothing to prove. So assume that c is referenced
in the services s1 , . . . , st that during the e-th epoch made the following dona-
tions to the protocol r1 , . . . , rt . While c is not referenced in the new services
st+1 , . . . , sk which made the following donations to the protocol rt+1 , . . . , rk
during the current epoch. When the number of profitable services increases,
so the number of profitable services passes from k to t + 1, t + 2, . . . , k, the
numerator of U CFc (e) remains r1 , . . . , rt while its denominator progressively
increases as follows
In other words, the denominator increases each time a new service makes
a donation to the protocol. Consequently, U CFc (e) decreases as more prof-
itable services non referencing c increases.
3. If U CFc (e) = 0 we have nothing to prove. Assume that the services s1 , . . . , sk
during the e-th epoch made respectively the following donations to the proto-
col r1 , . . . , rk . Assume that the number of profitable services not referencing
c decreases. Let t ≤ k.
So assume that c is referenced in the services s1 , . . . , st that during the e-th
epoch made the following donations to the protocol r1 , . . . , rt and assume
that the number of profitable services not referencing c decreases.
So firstly assume that c is referenced in the services s1 , . . . , st but not in
st+1 , . . . , sk . In this case:
Autonolas Tokenomics 29
r1 + · · · + rt
U CFc (e) = . (11)
r1 + · · · + rt + rt+1 + · · · + rk
Continuing in this way, until we have that c is referenced in all the services
s1 , . . . , st , st+1 , . . . , sk and so
r1 + · · · + rt + rt+1 + · · · + rk
U CFc (e) = . (13)
r1 + · · · + rt + rt+1 + · · · + rk
Since the numerator increases while the numerator remains constant, we have
that the fraction in the equation 11 is bigger than the one in equation 12
that in turn is bigger than the fraction in equation 13. So the statement is
proved.
4. Assume that c is referenced in the services s1 , . . . , st that during the e-th
epoch made the following donations to the protocol r1 , . . . , rt , while c is not
referenced in st+1 , . . . , sk . Assume further that the latter, during the e-th
epoch, made the following donations rt+1 , . . . , rk and that c′ is referenced in
all the services s1 , . . . , sk . Further
rt+1 + · · · + rk << r1 + · · · + rt ,
r1 + · · · + rt = 100(rt+1 + · · · + rk ).
Then
r1 + · · · + rt 100
U CFc (e) = = ∼ 0.99
r1 + · · · + rt + rt+1 + · · · + rk 101
while
U CFc′ (e) = 1.
30 Valory AG
δscki (e) · citcsij (e) = citcsij (e) · δscki (e) = δk (j)citcsij (e)
δsaki (e) · citasji (e) = citasji (e) · δsaki (e) = δk (j)citasji (e).
Proof. The proof is the same for component and agent, so we are going to only
consider the component ci . Replacing the definitions of U CFci (e) (cf. 5.1) and
Rcomp (e) (cf. 5.2) on the left-hand side of equation (14) yields:
Autonolas Tokenomics 31
The latter coincides with the definition given in equation (5), hence the result
follows.
9. When BR(t) ̸= 0 and K and D are such that f (K(e−1), D(e−1)) approaches
+∞, then DF (e) approaches
1
.
1 + ϵ(e)
10. When BR(t) ̸= 0, ϵ(e) > 0, and K(e − 1) and D(e − 1) approach 0, then
DF (e, t) approaches one.
f (K(e), D(e)) = d(k · K(e) + D(e)) < d(k · K(e) + D′ (e)) = f (K(e), D′ (e)).
3. The production function cannot have negative values because the worst case
is when the protocol receives no profits and there are no developers during
the epochs. So in the worst case, f (K, D) = 0.
4. Immediate from the definition of DF .
5. Easy to prove with basic knowledge of Calculus.
6. When BR(t) ̸= 0, then
1 1
DF (t, e) = max , .
1 + ϵ(e) 1 + f (K(e − 1), D(e − 1))/100
1
≤ 1.
1 + f (K(e − 1), D(e − 1))/100
Hence
1
DF (t, e) = .
1 + ϵ(e)
7. If ϵ(e) < −1, then
1
< 0.
1 + ϵ(e)
By 3. we have that f (K(e − 1), D(e − 1)) ≥ 0, so
1
0< ≤ 1.
1 + f (K(e − 1), D(e − 1))/100
Hence
1
DF (t, e) = .
1 + f (K(e − 1), D(e − 1))/100
8. Let ϵ(e). When f (K(e − 1), D(e − 1))/100 ≥ ϵ(e), then
1 1
≤ ,
1 + f (K(e − 1), D(e − 1))/100 1 + ϵ(e)
so
1
DF (t, e) = .
1 + ϵ(e)
If f (K(e − 1), D(e − 1))/100 ≤ ϵ(e),
1 1
≥
1 + f (K(e − 1), D(e − 1))/100 1 + ϵ(e)
So
1 1
DF (t, e) = ≥ .
1 + f (K(e − 1), D(e − 1))/100 1 + ϵ(e)
That is, once ϵ(e) it is fixed and it is bigger than 0, we have that DF (t, e)
can only be bigger than
1
.
1 + ϵ(e)
9. Since f (K, D) increases with K and D. Very large f (K(e − 1), D(e − 1))
corresponds to very small
1
.
1 + f (K(e − 1), D(e − 1))/100
1
10. Since f (K, D) approaches zero with K and D. Then 1+f (K(e−1),D(e−1))/100
approaches. Since for ϵ(e) > 0, we have that
1
< 1,
1 + ϵ(e)
then
1
DF (t, e) +
1 + f (K(e − 1), D(e − 1))/100
approaches 1.
Fig. 8. Gas units required to create a component with “number of splits” subcompo-
nents at the net of the gas units for one component. Limits per Ethereum block is 30
million units of gas.
D.2 Split-components
Let’s denote by component with n splits a component that has n−1 dependen-
cies (e.g. n − 1 subcomponents different from the component were referenced).
We consider that a component with n splits is malicious when it could have
been created with no dependency but instead was registered with n − 1 dummy
dependencies. Being its dependencies dummy, it is safe to assume that such
dependencies are only referenced only in services that reference the malicious
component itself and no other services reference any of them.
36 Valory AG
Let’s first consider the case where only one useful service, s, references the
malicious component c with n splits and that such a service references other dif-
ferent g components. Assume that thanks to s the protocol accrues as a donation
the amount r. From equation (5), we can deduce that the developer owner of c
receives the following reward
n
rew(n) = cf · r · , (16)
n+g
where we recall that cf ·r is the share of the donation r reserved to fund the useful
components in s. Note that, the reward approaches its maximum value cf · r
when the splits n is sufficiently larger than the number of extra components g.
Moreover, if a service reference a number of extra components g relatively larger
than the number of splits s, the reward of the malicious developer decreases. A
graphical representation of this is summarized in Fig. 9. Specifically, the blue
curve represents the percentage of the donation share cf · r that a malicious
component c with n splits can receive if the s references c and only one other
component different from c and its dependencies. Similarly, the purple curve
represents the percentage of the donation share cf · r that is given as a reward
to a malicious component c with n splits when the service referenced 100 extra
components. When the service s references 2 < m < 100 extra components, the
rewards curves sit between the blue and the purple ones.
Fig. 9 shows that when the number of splits n is sufficiently larger than
the number of extra components g, the rewards distribution approaches the
maximum percentage. However, also shows that when the number of extra com-
ponents increases the reward of the malicious developers decreases. E.g. when
the service references only one extra component, with just 10 splits a malicious
developer can accrue around 90% of the maximal reward cf · r, but, if at least
Autonolas Tokenomics 37
Let gmin and gmax be respectively the minimal and maximal value among the
extra components g1 , . . . , gt and let rmim and rmax be respectively the minimal
and maximal value among the donations r1 , . . . , rt . Then the following holds
n n
cf · t · · rmim ≤ rew(n) ≤ cf · t · · rmax , (18)
n + gmax n + gmin