Penumbra Guide

Penumbra is a fully shielded zone for the Cosmos ecosystem, allowing anyone to securely transact, stake, swap, or marketmake without broadcasting their personal information to the world.

This site contains documentation on how to use, deploy, and develop the Penumbra software. The description of the protocol itself can be found in the protocol specification, and the API documentation can be found here.

Test networks

Penumbra is a decentralized protocol, so Penumbra Labs is building in public, launching (and crashing) lots of work-in-progress testnets to allow community participation, engagement, and feedback.

Currently, Penumbra only has a command line client, pcli (pronounced “pickle-y”), which bundles all of the client components in one binary, and a chain-scanning daemon, pclientd, which runs just the view service, without spend capability. To get started with the Penumbra test network, all that’s required is to download and build pcli, as described in Installation.

The Penumbra node software is the Penumbra daemon, pd. This is an ABCI application, which must be driven by Tendermint, so a Penumbra full node consists of both a pd instance and a tendermint instance.

The basic architecture of Penumbra is as follows:

          ╭   ┌───────┐
  spending│   │custody│
capability│   │service│
          ╰   └───────┘
               ▲     │
               │tx   │auth
               │plan │data
               │     ▼
          ╭   ┌───────┐
   viewing│   │wallet │ tx submission
capability│   │logic  │────────┐
          │   └───────┘        │
          │    ▲               │
          │    │view private state
          │    │               │
          │    │               │
          │   ┌───────┐        │
          │   │view   │        │
          │   │service│        │
          ╰   └───────┘        │
               ▲               │
               │sync private state
               │               │
          ╭ ┌──┼───────────────┼──────┐
    public│ │  │     Penumbra Fullnode│
     chain│ │  │               │      │
      data│ │  │               ▼      │
          │ │ ┌──┐ app   ┌──────────┐ │
          │ │ │pd│◀─────▶│tendermint│ │
          │ │ └──┘ sync  └──────────┘ │
          │ │               ▲         │
          ╰ └───────────────┼─────────┘
                       ,'   │ `.
                  .───;     │consensus
                 ;          │sync
               .─┤          │   ├──.
             ,'             │       `.
            ;   Penumbra    │         :
            :   Network  ◀──┘         ;
             ╲                       ╱
              `.     `.     `.     ,'
                `───'  `───'  `───'

The custody service holds signing keys and is responsible for authorizing transaction plans. The view service holds viewing keys and scans the chain state. Wallet logic can query the view service to get information about what funds are available, submit a transaction plan to the custody service for signing, and then use the returned signatures to build the transaction and submit it.

As a shielded chain, Penumbra’s architecture is slightly different than a transparent chain, because user data such as account balances, transaction activity, etc., is not part of the public chain state. This means that clients need to synchronize with the chain to build a copy of the private user data they have access to. This logic is provided by the view service, which is bundled into pcli, but can also be run as a standalone pclientd daemon.

Modeling authorization as an (asynchronous) RPC to a custody service means that the client software is compatible with many different custody flows by default – an in-process “SoftHSM”, a hardware wallet with user intervention, a cluster of online threshold signers, an offline threshold signing process, etc.

Using pcli

This section describes how to use pcli, the command line client for Penumbra:

Penumbra is a private blockchain, so the public chain state does not reveal any private user data. By default, pcli includes a view service that synchronizes with the chain and scans with a viewing key. However, it’s also possible to run the view service as a standalone pclientd daemon:

Please submit any feedback and bug reports

Thank you for helping us test the Penumbra network! If you have any feedback, please let us know in the #testnet-feedback channel on our Discord. We would love to know about bugs, crashes, confusing error messages, or any of the many other things that inevitably won’t quite work yet. Have fun! :)

Diagnostics and Warnings

By default, pcli prints a warning message to the terminal, to be sure that people understand that this is unstable, unfinished, pre-release software. To disable this warning, export the PCLI_UNLEASH_DANGER environment variable.

Installing pcli

Installing the Rust toolchain

This requires that you install a recent stable version of the Rust compiler, installation instructions for which you can find here. Don’t forget to reload your shell so that cargo is available in your \$PATH!

pcli requires rustfmt as part of the build process — depending on your OS/install method for Rust, you may have to install that separately.

Installing build prerequisites


You may need to install some additional packages in order to build pcli, depending on your distribution. For a bare-bones Ubuntu installation, you can run:

sudo apt-get install build-essential pkg-config libssl-dev clang git-lfs

For a minimal Fedora/CentOS/RHEL image, you can run:

sudo dnf install openssl-devel clang git cargo rustfmt git-lfs


You may need to install the command-line developer tools if you have never done so:

xcode-select --install

You’ll also need to install Git LFS, which you can do via Homebrew:

brew install git-lfs

Cloning the repository

Once you have installed the above tools, you can clone the repository:

git clone

To build the version of pcli compatible with the current testnet, navigate to the penumbra folder, fetch the latest from the repository, and check out the latest tag for the current testnet:

cd penumbra && git fetch && git checkout 048-carme.1

Building the pcli client software

Then, build the pcli tool using cargo:

cargo build --release --bin pcli

Because you are building a work-in-progress version of the client, you may see compilation warnings, which you can safely ignore.

Generating a Wallet

On first installation of pcli, you will need to generate a fresh wallet to use with Penumbra. You should see something like this:

$ cargo run --quiet --release --bin pcli keys generate

Saving backup wallet to /home/$USER/.local/share/penumbra-testnet-archive/.../custody.json

Penumbra’s design automatically creates many (u64::MAX) publicly unlinkable addresses which all correspond to your own wallet. When you first created your wallet above, pcli initialized all of your wallet addresses, which you can view like this:

$ cargo run --quiet --release --bin pcli view address 0

Getting testnet tokens on the [Discord] in the #testnet-faucet channel

In order to use the testnet, it’s first necessary for you to get some testnet tokens. The current way to do this is to join our Discord and post your address in the #testnet-faucet channel. We’ll send your address some tokens on the testnet for you to send to your friends! :)

Just keep in mind: testnet tokens do not have monetary value, and in order to keep the signal-to-noise ratio high on the server, requests for tokens in other channels will be deleted without response. Please do not DM Penumbra Labs employees asking for testnet tokens; the correct venue is the dedicated channel.

Updating pcli

Make sure you’ve followed the installation steps. Then, to update to the latest testnet release:

cd penumbra && git fetch && git checkout 048-carme.1

Once again, build pcli with cargo:

cargo build --release --bin pcli

No wallet needs to be generated. Instead, keep one’s existing wallet and reset view data.

cargo run --quiet --release --bin pcli view reset

Viewing Balances

Once you’ve received your first tokens, you can scan the chain to import them into your local wallet (this may take a few minutes the first time you run it):

cargo run --quiet --release --bin pcli view sync

Syncing is performed automatically, but running the sync subcommand will ensure that the client state is synced to a recent state, so that future invocations of pcli commands don’t need to wait.

If someone sent you testnet assets, you should be able to see them now by running:

cargo run --quiet --release --bin pcli view balance

This will print a table of assets by balance in each. The balance view just shows asset amounts. To see more information about delegation tokens and the stake they represent, use

cargo run --quiet --release --bin pcli view staked

Sending Transactions

Now, for the fun part: sending transactions. If you have someone else’s testnet address, you can send them any amount of any asset you have.

First, use balance to find the amount of assets you have:

cargo run --release --bin pcli view balance

Second, if I wanted to send 10 penumbra tokens to my friend, I could do that like this (filling in their full address at the end):

cargo run --quiet --release --bin pcli tx send 10penumbra --to penumbrav2t...

Notice that asset amounts are typed amounts, specified without a space between the amount (10) and the asset name (penumbra). If you have the asset in your wallet to send, then so it shall be done!


In addition, to sending an asset, one may also stake penumbra tokens to validators.

Find a validator to stake to:

cargo run --release --bin pcli query validator list

Copy and paste the identity key of one of the validators to stake to, then construct the staking tx:

cargo run --release --bin pcli tx delegate 10penumbra --to penumbravalid...

To undelegate from a validator, use the pcli tx undelegate command, passing it the typed amount of delegation tokens you wish to undelegate. Wait a moment for the network to process the undelegation, then reclaim your funds:

cargo run --release --bin pcli tx undelegate-claim

Inspect the output; a message may instruct you to wait longer, for a new epoch. Check back and rerun the command later to add the previously delegated funds to your wallet.


Penumbra features on-chain governance similar to Cosmos Hub where anyone can submit proposals and both validators and delegators to vote on them. Penumbra’s governance model incorporates a single DAO account, into which anyone can freely deposit, but from which only a successful governance vote can spend. For details on using governance, see the governance section.

Swapping Assets

One of the most exciting features of Penumbra is that by using IBC (inter-blockchain communication) and our shielded pool design, any tokens can be exchanged in a private way.

NOTE: Since there’s not yet a way to open liquidity positions, we have implemented a stub Uniswap-V2-style constant-product market maker with some hardcoded liquidity for a few trading pairs: gm:gn, penumbra:gm, and penumbra:gn.

This allows testing the shielded pool integration, and will cause a floating exchange rate based on the volume of swaps occurring in each pair.

You can check the current reserves for a trading pair using the cpmm-reserves dex query:

cargo run --release --bin pcli -- q dex cpmm-reserves gm:penumbra

If you wanted to exchange 1 penumbra tokens for gm tokens, you could do so like so:

cargo run --release --bin pcli -- tx swap --into gm 1penumbra

This will handle generating the swap transaction and you’d soon have the market-rate equivalent of 1 penumbra in gm tokens returned to you, or the original investment of 1 penumbra tokens returned if there wasn’t enough liquidity available to perform the swap.


Penumbra features on-chain governance similar to Cosmos Hub, with the simplification that there are only 3 kinds of vote: yes, no, and abstain.

Quick Start

There’s a lot you can do with the governance system in Penumbra. If you have a particular intention in mind, here are some quick links:

Getting Proposal Information

To see information about the currently active proposals, including your own, use the pcli query proposal subcommand.

To list all the active proposals by their ID, use:

cargo run --release --bin pcli query governance list-proposals

Other proposal query commands all follow the form:

cargo run --release --bin pcli query governance proposal [PROPOSAL_ID] [QUERY]

These are the queries currently defined:

  • definition gets the details of a proposal, as the submitted JSON;
  • state gets information about the current state of a proposal (voting, withdrawn, or finished, along with the reason for withdrawal if any, and the outcome of finished proposals);
  • period gets the voting start and end block heights of a proposal;
  • tally gets the current tally of a proposal’s votes, as a total across all validators, and broken down by each validator’s votes and the total votes of their delegators.

Voting On A Proposal

Validators and delegators may both vote on proposals. Validator votes are public and attributable to that validator; delegator votes are anonymous, revealing only the voting power used in the vote, and the validator which the voting delegator had delegated to. Neither validators nor delegators can change their votes after they have voted.

Voting As A Delegator

If you had staked delegation tokens to one or more active validators when a proposal started, you can vote on it using the tx vote subcommand of pcli. For example, if you wanted to vote “yes” on proposal 1, you would do:

cargo run --release --bin pcli tx vote yes --on 1

When you vote as a delegator (but not when you vote as a validator), you will receive commemorative voted_on_N tokens, where N is the proposal ID, proportionate to the weight of your vote. Think of these as the cryptocurrency equivalent of the “I voted!” stickers you may have received when voting in real life at your polling place.

Voting As A Validator

If you are a validator who was active when the proposal started, you can vote on it using the validator vote subcommand of pcli. For example, if you wanted to vote “yes” on proposal 1, you would do:

cargo run --release --bin pcli validator vote yes --on 1

Eligibility And Voting Power

Only validators who were active at the time the proposal started voting may vote on proposals. Only delegators who had staked delegation tokens to active validators at the time the proposal started voting may vote on proposals.

A validator’s voting power is equal to their voting power at the time a proposal started voting, and a delegator’s voting power is equal to the unbonded staking token value (i.e. the value in penumbra) of the delegation tokens they had staked to an active validator at the time the proposal started voting. When a delegator votes, their voting power is subtracted from the voting power of the validator(s) to whom they had staked delegation notes at the time of the proposal start, and their stake-weighted vote is added to the total of the votes: in other words, validators vote on behalf of their delegators, but delegators may override their portion of their validator’s vote.

Authoring A Proposal

Anyone can submit a new governance proposal for voting by escrowing a proposal deposit, which will be held until the end of the proposal’s voting period. Penumbra’s governance system discourages proposal spam with a slashing mechanism: proposals which receive more than a high threshold of no votes have their deposit burned. At present, the slashing threshold is 80%. If the proposal is not slashed (but regardless of whether it passes or fails), the deposit will then be returned to the proposer at the end of voting.

From the proposer’s point of view, the lifecycle of a proposal begins when it is submitted and ends when it the deposit is claimed. During the voting period, the proposer may also optionally withdraw the proposal, which prevents it from passing, but does not prevent it from being slashed. This is usually used when a proposal has been superceded by a revised alternative.

In the above, rounded grey boxes are actions submitted by the proposal author, rectangular colored boxes are the state of the proposal on chain, and colored circles are outcomes of voting.

Kinds Of Proposal

There are 4 kinds of governance proposal on Penumbra: signaling, emergency, parameter change, and DAO spend.

Signaling Proposals

Signaling proposals are meant to signal community consensus about something. They do not have a mechanized effect on the chain when passed; they merely indicate that the community agrees about something.

This kind of proposal is often used to agree on code changes; as such, an optional commit field may be included to specify these changes.

Emergency Proposals

Emergency proposals are meant for when immediate action is required to address a crisis, and conclude early as soon as a 2/3 majority of all active voting power votes yes.

Emergency proposals have the power to optionally halt the chain when passed. If this occurs, off-chain coordination between validators will be required to restart the chain.

Parameter Change Proposals

Parameter change proposals alter the chain parameters when they are passed. Chain parameters specify things like the base staking reward rate, the amount of penalty applied when slashing, and other properties that determine how the chain behaves. Many of these can be changed by parameter change proposals, but some cannot, and instead would require a chain halt and upgrade.

A parameter change proposal specifies both the old and the new parameters. If the current set of parameters at the time the proposal passes are an exact match for the old parameters specified in the proposal, the entire set of parameters is immediately set to the new parameters; otherwise, nothing happens. This is to prevent two simultaneous parameter change proposals from overwriting each others’ changes or merging with one another into an undesired state. Almost always, the set of old parameters should be the current parameters at the time the proposal is submitted.

DAO Spend Proposals

DAO spend proposals submit a transaction plan which may spend funds from the DAO if passed.

DAO spend transactions have exclusive capability to use two special actions which are not allowed in directly submitted user transactions: DaoSpend and DaoOutput. These actions, respectively, spend funds from the DAO, and mint funds transparently to an output address (unlike regular output actions, which are shielded). DAO spend transactions are unable to use regular shielded outputs, spend funds from any source other than the DAO itself, perform swaps, or submit, withdraw, or claim governance proposals.

Submitting A Proposal

To submit a proposal, first generate a proposal template for the kind of proposal you want to submit. For example, suppose we want to create a signaling proposal:

cargo run --release --bin pcli tx proposal template signaling --file proposal.toml

This outputs a TOML template for the proposal to the file proposal.toml, where you can edit the details to match what you’d like to submit. The template will contain relevant default fields for you to fill in, as well as a proposal ID, automatically set to the next proposal ID at the time you generated the template. If someone else submits a proposal before you’re ready to upload yours, you may need to increment this ID, because it must be the sequentially next proposal ID at the time the proposal is submitted to the chain.

Once you’re ready to submit the proposal, you can submit it. Note that you do not have to explicitly specify the proposal deposit in this action; it is determined automatically based on the chain parameters.

cargo run --release --bin pcli tx proposal submit --file proposal.toml

The proposal deposit will be immediately escrowed and the proposal voting period will start in the very next block. As the proposer, you will receive a proposal deposit NFT which can be redeemed for the proposal deposit after voting concludes, provided the proposal is not slashed. This NFT has denomination proposal_N_deposit, where N is the ID of your proposal. Note that whoever holds this NFT has exclusive control of the proposal: they can withdraw it or claim the deposit.

Making A DAO Spend Transaction Plan

In order to submit a DAO spend proposal, it is necessary to create a transaction plan. At present, the only way to specify this is to provide a rather human-unfriendly JSON-formatted transaction plan, because there is no stable human-readable representation for a transaction plan at present. This will change in the future as better tooling is developed.

For now, here is a template for a transaction plan that withdraws 100 penumbra from the DAO and sends it to a specified address (in this case, the address of the author of this document):

    "fee": { "amount": { "lo": 0, "hi": 0 } },
    "actions": [
        { "daoSpend": { "value": {
            "amount": { "lo": 100000000, "hi": 0 },
            "assetId": { "inner": "KeqcLzNx9qSH5+lcJHBB9KNW+YPrBk5dKzvPMiypahA=" }
        } } },
        { "daoOutput": {
            "value": {
                "amount": { "lo": 100000000, "hi": 0 },
                "assetId": { "inner": "KeqcLzNx9qSH5+lcJHBB9KNW+YPrBk5dKzvPMiypahA=" }
            "address": {
                "inner": "vzZ60xfMPPwewTiSb08jk5OdUjc0BhQ7IXLgHAayJoi5mvmlnTpqFuaPU2hCBhwaEwO2c03tBbN/GVh0+CajAjYBmBq3yHAbzNJCnZS8jUs="
        } }

Note that the asset ID and address are specified not in the usual bech32 formats you are used to seeing, but in base64. To get your address in this format, use pcli view address 0 --base64.

To template a DAO spend proposal using a JSON transaction plan, use the pcli tx proposal template dao-spend --transaction-plan <FILENAME>.json, which will take the transaction plan and include it in the generated proposal template. If no plan is specified, the transaction plan will be the empty transaction which does nothing when executed.

Withdrawing A Proposal

If you want to withdraw a proposal that you have made (perhaps because a better proposal has come to community consensus), you can do so before voting concludes. Note that this does not make you immune to losing your deposit by slashing, as withdrawn proposals can still be voted on and slashed.

cargo run --release --bin pcli \
    tx proposal withdraw 0 \
    --reason "some human-readable reason for withdrawal"

When you withdraw a proposal, you consume your proposal deposit NFT, and produce a new proposal unbonding deposit NFT, which has the denomination proposal_N_unbonding_deposit, where N is the proposal ID. This, like the proposal deposit NFT, can be used to redeem the deposit at the end of voting, provided the proposal is not slashed.

Claiming A Proposal Deposit

Regardless of whether you have or have not withdrawn your proposal, once voting on the proposal concludes, you can claim your proposal deposit using the tx proposal deposit-claim subcommand of pcli. For example, if you wanted to claim the deposit for a concluded proposal number 1, you could say:

cargo run --release --bin pcli tx proposal deposit-claim 1

This will consume your proposal deposit NFT (either the original or the one you received after withdrawing the proposal) and send you back one of three different proposal result NFTs, depending on the result of the vote: proposal_N_passed, proposal_N_failed or proposal_N_slashed. If the proposal was not slashed (that is, it passed or failed), this action will also produce the original proposal deposit. Note that you can claim a slashed proposal: you will receive the slashed proposal result NFT, but you will not receive the original proposal deposit.

Contributing To The DAO

Anyone can contribute any amount of any denomination to the Penumbra DAO. To do this, use the command pcli tx dao-deposit, like so:

cargo run --release --bin pcli tx dao-deposit 100penumbra

Funds contributed to the DAO cannot be withdrawn except by a successful DAO spend governance proposal.

To query the current DAO balance, use pcli query dao balance with the base denomination of an asset or its asset ID (display denominations are not currently accepted). For example:

cargo run --release --bin pcli query dao balance upenumbra

DAO spend proposals are only accepted for voting if they would not overdraw the current funds in the DAO at the time the proposal is submitted, so it’s worth checking this information before submitting such a proposal.

Sending Validator Funding Streams To The DAO

A validator may non-custodially send funds to the DAO, similarly to any other funding stream. To do this, add a [[funding_stream]] section to your validator definition TOML file that declares the DAO as a recipient for a funding stream. For example, your definition might look like this:

sequence_number = 0
enabled = true
name = "My Validator"
website = ""
description = "An example validator"
identity_key = "penumbravalid1s6kgjgnphs99udwvyplwceh7phwt95dyn849je0jl0nptw78lcqqvcd65j"
governance_key = "penumbragovern1s6kgjgnphs99udwvyplwceh7phwt95dyn849je0jl0nptw78lcqqhknap5"

type = "tendermint/PubKeyEd25519"
value = "tDk3/k8zjEyDQjQC1jUyv8nJ1cC1B/MgrDzeWvBTGDM="

# Send a 1% commission to this address:
recipient = "penumbrav2t1hum845ches70c8kp8zfx7nerjwfe653hxsrpgwepwtspcp4jy6ytnxhe5kwn56sku684x6zzqcwp5ycrkee5mmg9kdl3jkr5lqn2xq3kqxvp4d7gwqdue5jznk2ter2t66mk4n"
rate_bps = 100

# Send another 1% commission to the DAO:
recipient = "DAO"
rate_bps = 100

Using pcli with pclientd

First, export a viewing key from pcli:

pcli keys export full-viewing-key

Next, use the FVK it prints to initialize the pclientd state:

pclientd init FVK_STRING

The location of the pclientd state can be changed with the -s parameter. Finally, run

pclientd start

to start the view server, and invoke pcli with

pcli -v

to use it instead of an in-process view service.

WARNING: the view service does not currently use transport encryption, so it should not be used over a public network.

Using pd

This section describes how to build and run pd, the node implementation for Penumbra:

Building pd

The node software pd is part of the same repository as pcli, so follow those instructions to clone the repo and install dependencies.

To build pd, run

cargo build --release --bin pd

Because you are building a work-in-progress version of the node software, you may see compilation warnings, which you can safely ignore.

Installing Tendermint

You’ll need to have Tendermint installed on your system to join your node to the testnet.

NOTE: Previous versions of Penumbra used Tendermint 0.35, which has now been officially deprecated by the Tendermint Council. We have now rolled back to v0.34. Do not use Tendermint 0.35, which will no longer work with pd. that can prevent nodes from staying online.

Follow Tendermint’s installation instructions, but before you start compiling, make sure you are compiling version v0.34.23.

git checkout v0.34.23

Joining a Testnet

We provide instructions for running both fullnode deployments and validator deployments. A fullnode will sync with the network but will not have any voting power, and will not be eligible for staking or funding stream rewards. For more information on what a fullnode is, see the Tendermint documentation.

A regular validator will participate in voting and rewards, if it becomes part of the consensus set. Of course, these rewards, like all other testnet tokens, have no value.

Joining as a fullnode

To join a testnet as a fullnode, check out the tag for the current testnet, run pd testnet join to generate configs, then use those configs to run pd and tendermint. In more detail:

Resetting state

First, reset the testnet data from any prior testnet you may have joined:

cargo run --bin pd --release -- testnet unsafe-reset-all

This will delete the entire testnet data directory.

Generating configs

Next, generate a set of configs for the current testnet:

cargo run --bin pd --release -- testnet join --external-address IP_ADDRESS --moniker MY_NODE_NAME

where IP_ADDRESS (like is the public IP address of the node you’re running, and MY_NODE_NAME is a moniker identifying your node. Other peers will try to connect to your node over port 26656/TCP.

If your node is behind a firewall or not publicly routable for some other reason, skip the --external-address flag, so that other peers won’t try to connect to it. You can also skip the --moniker flag to use a randomized moniker instead of selecting one.

This command fetches the genesis file for the current testnet, and writes configs to a testnet data directory (by default, ~/.penumbra/testnet_data). If any data exists in the testnet data directory, this command will fail. See the section above on resetting node state.

Running pd and tendermint

Next, run pd with the --home parameter pointed at the correct part of the testnet data directory. It’s useful to set the RUST_LOG environment variable to get information about what it’s doing:

export RUST_LOG="warn,pd=debug,penumbra=debug" # or some other logging level
cargo run --bin pd --release -- start --home ~/.penumbra/testnet_data/node0/pd

Then (perhaps in another terminal), run Tendermint, also specifying --home:

tendermint start --home ~/.penumbra/testnet_data/node0/tendermint

Alternatively, pd and tendermint can be orchestrated with docker-compose:

cd deployments/compose/
docker-compose -f docker-compose.yml -f up -d --build

or via systemd:

cd deployments/systemd/
sudo cp *.service /etc/systemd/system/
# edit service files to customize for your system
sudo systemctl daemon-reload
sudo systemctl restart penumbra tendermint

Joining as a validator

After starting your node, as above, you should now be participating in the network as a fullnode. However your validator won’t be visible to the chain yet, as the definition hasn’t been uploaded.

Validator Definitions (Penumbra)

A validator definition contains fields defining metadata regarding your validator as well as funding streams, which are Penumbra’s analogue to validator commissions.

The root of a validator’s identity is their identity key. Currently, pcli reuses the spend authorization key in whatever wallet is active as the validator’s identity key. This key is used to sign validator definitions that update the configuration for a validator.

Creating a template definition

To create a template configuration, use pcli validator definition template:

$ cargo run --release --bin pcli -- validator definition template \
    --tendermint-validator-keyfile ~/.penumbra/testnet_data/node0/tendermint/config/priv_validator_key.json \
    --file validator.toml
$ cat validator.toml
# This is a template for a validator definition.
# The identity_key and governance_key fields are auto-filled with values derived
# from this wallet's account.
# You should fill in the name, website, and description fields.
# By default, validators are disabled, and cannot be delegated to. To change
# this, set `enabled = true`.
# Every time you upload a new validator config, you'll need to increment the
# `sequence_number`.

sequence_number = 0
enabled = false
name = ''
website = ''
description = ''
identity_key = 'penumbravalid1kqrecmvwcc75rvg9arhl0apsggtuannqphxhlzl34vfamp4ukg9q87ejej'
governance_key = 'penumbragovern1kqrecmvwcc75rvg9arhl0apsggtuannqphxhlzl34vfamp4ukg9qus84v5'

type = 'tendermint/PubKeyEd25519'
value = 'HDmm2FmJhLHxaKPnP5Fw3tC1DtlBx8ETgTL35UF+p6w='

recipient = 'penumbrav2t1cntf73e36y3um4zmqm4j0zar3jyxvyfqxywwg5q6fjxzhe28qttppmcww2kunetdp3q2zywcakwv6tzxdnaa3sqymll2gzq6zqhr5p0v7fnfdaghrr2ru2uw78nkeyt49uf49q'
rate_bps = 100

recipient = "DAO"
rate_bps = 100

and adjust the data like the name, website, description, etc as desired.

The enabled field can be used to enable/disable your validator without facing slashing penalties. Disabled validators can not appear in the active validator set and are ineligible for rewards.

This is useful if, for example, you know your validator will not be online for a period of time, and you want to avoid an uptime violation penalty. If you are uploading your validator for the first time, you will likely want to start with it disabled until your Tendermint & pd instances have caught up to the consensus block height.

Note that by default the enabled field is set to false and will need to be enabled in order to activate one’s validator.

In the default template, there is a funding stream declared to contribute funds to the DAO. This is not required, and may be altered or removed if you wish.

Setting the consensus key

In the command above, the --tendermint-validator-keyfile flag was used to instruct pcli to import the consensus key for the Tendermint identity. This works well if pcli and pd are used on the same machine. If you are running them in separate environments, you can omit the flag, and pd will generate a random key in the template. You must then manually update the consensus_key. You can get the correct value for consensus_key from your tendermint configs:

$ grep -A3 pub_key ~/.penumbra/testnet_data/node0/tendermint/config/priv_validator_key.json
  "pub_key": {
    "type": "tendermint/PubKeyEd25519",
    "value": "Fodjg0m1kF/6uzcAZpRcLJswGf3EeNShLP2A+UCz8lw="

Copy the string in the value field and paste that into your validator.toml, as the value field under the [consensus_key] heading.

Configuring funding streams

Unlike the Cosmos SDK, which has validators specify a commission percentage that goes to the validator, Penumbra uses funding streams, a list of pairs of commission amounts and addresses. This design allows validators to dedicate portions of their commission non-custodially – for instance, a validator could declare some amount of commission to cover their operating costs, and another that would be sent to an address controlled by a DAO.

Uploading a definition

After setting up metadata, funding streams, and the correct consensus key in your validator.toml, you can upload it to the chain:

cargo run --release --bin pcli -- validator definition upload --file validator.toml

And verify that it’s known to the chain:

cargo run --release --bin pcli -- query validator list -i

However your validator doesn’t have anything delegated to it and will remain in an Inactive state until it receives enough delegations to place it in the active set of validators.

Delegating to your validator

First find your validator’s identity key:

cargo run --release --bin pcli -- validator identity

And then delegate some amount of penumbra to it:

cargo run --release --bin pcli -- tx delegate 1penumbra --to penumbravalid1g2huds8klwypzczfgx67j7zp6ntq2m5fxmctkf7ja96zn49d6s9qz72hu3

You should then see your balance of penumbra decreased and that you have received some amount of delegation tokens for your validator:

cargo run --release --bin pcli view balance

Voting power will be calculated on the next epoch transition after your delegation takes place. Assuming that your delegation was enough to place your validator in the top N validators by voting power, it should appear in the validator list as Active after the next epoch transition. The epoch duration and the active validator limit are chain parameters, and will vary by deployment. You can find the values in use for the current chain in its genesis.json file.

Updating your validator

First fetch your existing validator definition from the chain:

cargo run --release --bin pcli -- validator definition fetch --file validator.toml

Then make any changes desired and make sure to increase by sequence_number by at least 1! The sequence_number is a unique, increasing identifier for the version of the validator definition.

After updating the validator definition you can upload it again to update your validator metadata on-chain:

cargo run --release --bin pcli -- validator definition upload --file validator.toml


This section is for developers working on pd itself.

Devnet Quickstart

This page describes a quickstart method for running pd+tendermint to test changes during development.

To start, you’ll need to install Tendermint v0.34.

Generating configs

To generate a clean set of configs, run

cargo run --release --bin pd -- testnet generate

This will write configs to ~/.penumbra/testnet_data/.

Running pd

You’ll probably want to set RUST_LOG. Here’s one suggestion:

export RUST_LOG="warn,pd=debug,penumbra=debug,jmt=info"

To run pd, run

cargo run --release --bin pd -- start  --home ~/.penumbra/testnet_data/node0/pd

This will start but won’t do anything yet, because Tendermint isn’t running.

Running tendermint

To run Tendermint, run

tendermint --home ~/.penumbra/testnet_data/node0/tendermint/ start

in another terminal window.

Running pcli

To interact with the chain, first do

cargo run --release --bin pcli -- view reset

and then pass the -n flag to any commands you run to point pcli at your local node, e.g.,

cargo run --bin pcli -- -n view balance

By default, pd testnet generate uses the latest snapshot of the Discord’s faucet channel, so if you posted your address there more than a week or two ago, you should already have an allocation in your local devnet.

If not, reset the state as below, and edit the genesis.json to add your address.

Resetting and restarting

After making changes, you may want to reset and restart the devnet:

cargo run --release --bin pd -- testnet unsafe-reset-all

You’ll probably also want to reset your wallet state:

cargo run --release --bin pcli -- view reset

At this point you’re ready to generate new configs, and restart both pd and tendermint. The order they’re started in doesn’t particularly matter for correctness, because tendermint will retry connecting to the ABCI server until it succeeds.

Optional: running smoke-tests

Once you have a working devnet running, you should be able to run the smoke tests successfully. This can be useful if you are looking to contribute to Penumbra, or if you need to check that your setup is correct.

To run the smoke tests:

  1. Make sure you have a devnet running (see previous steps)
  2. Run integration tests:
PENUMBRA_NODE_HOSTNAME= PCLI_UNLEASH_DANGER=yes cargo test --package pcli -- --ignored --test-threads 1

SQLite compilation setup

The view server uses SQLite via sqlx as its backing store. The type-safe query macros require compile-time information about the database schemas. Normally, this information is cached in the crate’s sqlx-data.json, and nothing extra is required to build.

However, when editing the view server’s database code, it’s necessary to work with a development database:

  1. You’ll need sqlx-cli installed with the correct features: cargo install sqlx-cli --features sqlite

  2. The database structure is defined in the migrations/ directory of the view crate.

  3. Set the DATABASE_URL environment variable to point to the SQLite location. For instance,

    export DATABASE_URL="sqlite:///tmp/pclientd-dev-db.sqlite"

    will set the shell environment variable to the same one set in the project’s .vscode/settings.json.

  4. From the view directory, run cargo sqlx database setup to create the database and run migrations.

  5. From the view directory, run cargo sqlx prepare -- --lib to regenerate the sqlx-data.json file that allows offline compilation.

Building documentation

The protocol docs and the guide (this document) are built using mdBook and auto-deployed on pushes to main. To build locally:

  1. Install the requirements: cargo install mdbook mdbook-katex mdbook-mermaid
  2. Run mdbook serve from docs/protocol (for the protocol spec) or from docs/guide (for this document).

The Rust API docs can be built with ./deployments/scripts/rust-docs. The landing page, the top-level index.html, is handled as a special case. If you added new crates by appending a -p <crate_name> to the rust-docs script, then you must rebuild the index page via:

REGENERATE_INDEX=1 ./deployments/scripts/rust-docs

CI will automatically rebuild all our docs on merges into main, except for the Rust doc index.html, which must be updated manually.

Maintaining protobuf specs

The Penumbra project dynamically generates code for interfacing with gRPC. The following locations within the repository are relevant:

  • proto/proto/penumbra/**/*.proto, the developer-authored spec files
  • proto/src/gen/*.rs, the generated Rust code files
  • tools/proto-compiler/, the build logic for generated the Rust code files

We use buf to auto-publish the protobuf schemas at, and to generate Go and Typescript packages. The Rust code files are generated with our own tooling, located at tools/proto-compiler.

Our custom tooling for generating the Rust files will also shape the Serde implementations of the derived Rust types to have more favorable JSON output (such as rendering addresses as Bech32-encoded strings).

Installing protobuf

Obtain the most recent pre-compiled binary from the protoc website. After installing, run protoc --version and confirm you’re running at least 3.21.8 (or newer). Don’t install protoc from package managers such as apt, as those versions are often outdated, and will not work with Penumbra.

Building protobuf

Switch to the proto-compiler directory and run the tool:

cd tools/proto-compiler
cargo run

Then run git status to determine whether any changes were made. The build process is deterministic, so regenerating multiple times from the same source files should not change the output. A possible exception to this rule is if prost makes a superficial change to the output that isn’t substantive.

If the generated output would change in any way, CI will fail, prompting the developer to commit the changes.


Metrics are an important part of observability, allowing us to understand what the Penumbra software is doing.

Viewing Metrics

TODO: add details on how to use Grafana:

  • link to for dashboard on current testnet;
  • instructions on how to run Grafana + Prometheus for local dev setup (ideally this could work without requiring that pd itself is Dockerized, since local development is often more convenient outside of docker);
  • instructions on how to commit dashboards back to the repo.

Adding Metrics

We use a common structure for organizing metrics code throughout the penumbra workspace. Each crate that uses metrics has a top-level metrics module, which is private to the crate. That module contains:

  • a re-export of the entire metrics crate: pub use metrics::*;
  • &'static str constants for every metrics key used by the crate;
  • a pub fn register_metrics() that registers and describes all of the metrics used by the crate;

Finally, the register_metrics function is publicly re-exported from the crate root.

The only part of this structure visible outside the crate is the register_metrics function in the crate root, allowing users of the library to register and describe the metrics it uses on startup.

Internally to the crate, all metrics keys are in one place, rather than being scattered across the codebase, so it’s easy to see what metrics there are. Because the metrics module re-exports the contents of the metrics crate, doing use crate::metrics; is effectively a way to monkey-patch the crate-specific constants into the metrics crate, allowing code like:

fn main() {
    "kind" => "new",
    "code" => "1"

The metrics keys themselves should:

  • follow the Prometheus metrics naming guidelines
  • have an initial prefix of the form penumbra_CRATE, e.g., penumbra_stake, penumbra_pd, etc;
  • have some following module prefix that makes sense relative to the other metrics in the crate.

For instance:

fn main() {
pub const MEMPOOL_CHECKTX_TOTAL: &str = "penumbra_pd_mempool_checktx_total";

Backing up Grafana

After being changed, Grafana dashboards should be backed up to the repository for posterity and redeployment.

Grafana has an import/export feature that we use for maintaining our dashboards.

  1. Export the dashboard as JSON with the default settings
  2. Rename the JSON file and copy into the repo (config/grafana/dashboards/)
  3. Use the import function in the UI to update all deployments


This page links to various resources that are helpful for working with and understanding Penumbra:


The primary communication hub is our Discord; click the link to join the discussion there.


Documentation on how to use pcli, how to run pd, and how to do development can be found at

Protocol Specification

The protocol specification is rendered at

API documentation

The API documentation is rendered at

Protobuf documentation

The protobuf documentation is rendered at

Talks and presentations

These talks were given at various conferences and events, describing different aspects of the Penumbra ecosystem.