trader-quickstart

A quickstart for the trader agent for AI prediction markets on Gnosis at https://github.com/valory-xyz/trader

Compatible Systems

• Windows 10/11: WSL2 / Git BASH
• Mac ARM / Intel
• Linux
• Raspberry Pi 4

System Requirements

Ensure your machine satisfies the requirements:

• Python ==3.10
• Poetry >=1.4.0
• Docker Engine <25.0.0
• Docker Compose

Resource Requirements

• You need xDAI on Gnosis Chain in one of your wallets.
• You need an RPC for your agent instance. We recommend Nodies RPC.
• (From release v0.16.0 onwards) You will need a Subgraph API key that can be obtained at The Graph.

Run the Service

For Non-Stakers

Clone this repository locally and execute:

chmod +x run_service.sh
./run_service.sh

Answer 1) No staking when prompted:

Please, select your staking program preference

For Stakers

⚠️ Warning
The code within this repository is provided without any warranties. It is important to note that the code has not been audited for potential security vulnerabilities. Using this code could potentially lead to loss of funds, compromised data, or asset risk. Exercise caution and use this code at your own risk. Please refer to the LICENSE file for details about the terms and conditions.

Each staking program has different OLAS requirements. The script will check that your owner address meets the minimum required OLAS on the Gnosis Chain.

Clone this repository locally and execute:

chmod +x run_service.sh
./run_service.sh

Select your preferred staking program when prompted:

Please, select your staking program preference
----------------------------------------------
1) No staking
   Your Olas Predict agent will still actively participate in prediction
   markets, but it will not be staked within any staking program.

2) Quickstart Beta - Hobbyist
   The Quickstart Beta - Hobbyist staking contract offers 100 slots for
   operators running Olas Predict agents with the quickstart. It is designed as
   a step up from Coastal Staker Expeditions, requiring 100 OLAS for staking.
   The rewards are also more attractive than with Coastal Staker Expeditions.

3) Quickstart Beta - Expert
   The Quickstart Beta - Expert staking contract offers 20 slots for operators
   running Olas Predict agents with the quickstart. It is designed for
   professional agent operators, requiring 1000 OLAS for staking. The rewards
   are proportional to the Quickstart Beta - Hobbyist.

...

Find below a diagram of the possible status a service can be in the staking program:

Services can become staked by invoking the stake() contract method, where service parameters and deposit amounts are verified. Staked services can call the checkpoint() method at regular intervals, ensuring liveness checks and calculating staking rewards. In case a service remains inactive beyond the specified maxAllowedInactivity time, it faces eviction from the staking program, ceasing to accrue additional rewards. Staked or evicted services can be unstaked by calling the unstake() contract method. They can do so after minStakingDuration has passed or if no more staking rewards are available.

Notes:

• Staking is currently in a testing phase, so the number of trader agents that can be staked might be limited.

• Services are evicted after accumulating 2 consecutive checkpoints without meeting the activity threshold.

• Currently, the minimum staking time is approximately 3 days. In particular, a service cannot be unstaked during the minimum staking period.

• Once a staking program is selected, you can reset your preference by stopping your agent by running ./stop_service.sh and then running the command

  ./reset_staking.sh

  Keep in mind that your service must stay for minStakingDuration in a staking program (typically 3 days) before you can change to a new program.

Service is Running

Once the command has completed, i.e. the service is running, you can see the live logs with:

docker logs trader_abci_0 --follow

To stop your agent, use:

./stop_service.sh

Claim accrued OLAS staking rewards

If your service is staked, you can claim accrued OLAS staking rewards through the script

./claim_staking_rewards.sh

The accrued OLAS will be transferred to your service Safe without having to unstake your service.

Backups

Agent runners are recommended to create a backup of the relevant secret key material.

Skip user prompts

You can optionally pass --attended=false or export the environment variable ATTENDED=false to skip asking for inputs from the user. Note: In this case, if the service is staked, then it will not update the on-chain service, to avoid unstaking. If you choose a staking contract with different parameters (e.g., different bond), then you have to execute the attended mode of the script to update the on-chain service.

Observe your agents

1. Use the trades command to display information about placed trades by a given address:

   cd trader; poetry run python ../trades.py --creator YOUR_SAFE_ADDRESS; cd ..

   Or restrict the search to specific dates by defining the "from" and "to" dates:

   cd trader; poetry run python ../trades.py --creator YOUR_SAFE_ADDRESS --from-date 2023-08-15:03:50:00 --to-date 2023-08-20:13:45:00; cd ..

2. Use the report command to display a summary of the service status:

   cd trader; poetry run python ../report.py; cd ..

3. Use this command to investigate your agent's logs:

   cd trader; poetry run autonomy analyse logs --from-dir trader_service/abci_build/persistent_data/logs/ --agent aea_0 --reset-db; cd ..

   For example, inspect the state transitions using this command:

   cd trader; poetry run autonomy analyse logs --from-dir trader_service/abci_build/persistent_data/logs/ --agent aea_0 --fsm --reset-db; cd ..

   This will output the different state transitions of your agent per period, for example:

   

   For more options on the above command run:

   cd trader; poetry run autonomy analyse logs --help; cd ..

   or take a look at the command documentation.

Update between versions

Simply pull the latest script:

git pull origin

Remove the existing trader folder:

rm -rf trader

Then continue above with "Run the script".

Change the password of your key files

⚠️ Warning
The code within this repository is provided without any warranties. It is important to note that the code has not been audited for potential security vulnerabilities.

If you are updating the password for your key files, it is strongly advised to create a backup of the old configuration (located in the ./trader_runner folder) before proceeding. This backup should be retained until you can verify that the changes are functioning as expected. For instance, run the service multiple times to ensure there are no issues with the new password before discarding the backup.

If you have started you script specifying a password to protect your key files, you can change it by running the following command:

cd trader; poetry run python ../scripts/change_keys_json_password.py ../.trader_runner --current_password <current_password> --new_password <new_password>; cd ..

This will change the password in the following files:

• .trader_runner/keys.json
• .trader_runner/operator_keys.json
• .trader_runner/agent_pkey.txt
• .trader_runner/operator_pkey.txt

If your key files are not encrypted, you must not use the --current-password argument. If you want to remove the password protection of your key files, you must not specify the --new-password argument.

Advice for Mac users

In Docker Desktop make sure that in Settings -> Advanced the following boxes are ticked

Advice for Windows users using Git BASH

We provide some hints to have your Windows system ready to run the agent. The instructions below have been tested in Windows 11.

Execute the following steps in a PowerShell terminal:

1. Install Git and Git Bash:

   winget install --id Git.Git -e --source winget

2. Install Python 3.10:

   winget install Python.Python.3.10

3. Close and re-open the PowerShell terminal.

4. Install Poetry:

   curl.exe -sSL https://install.python-poetry.org | python -

5. Add Poetry to your user's path:

   $existingUserPath = (Get-Item -Path HKCU:\Environment).GetValue("PATH", $null, "DoNotExpandEnvironmentNames")
   
   $newUserPath = "$existingUserPath;$Env:APPDATA\Python\Scripts"
   
   [System.Environment]::SetEnvironmentVariable("Path", $newUserPath, "User")

6. Install Docker Desktop:

   winget install -e --id Docker.DockerDesktop

7. Log out of your Windows session and then log back in.

8. Open Docker Desktop and leave it opened in the background.

Now, open a Git Bash terminal and follow the instructions in the "Run the script" section as well as the subsequent sections. You might need to install Microsoft Visual C++ 14.0 or greater.

Advanced usage

This chapter is for advanced users who want to further customize the trader agent's behaviour without changing the underlying trading logic.

Policy weights

This script automatically sets some default weights to the agent's policy as a warm start. to help convergence and improve tool selection. These data were obtained after many days of running the service and are set here. As a result, the current weights are always deleted and replaced by this strategy which is considered to boost the initial performance of the service.

However, you may have found better performing policy weights and would like to remove this logic. It can easily be done, by removing this method call, here, in order to set your own custom warm start. Setting your own custom weights can be done by editing the corresponding files in .trader_runner. Moreover, you may store your current policy as a backup before editing those files, using the following set of commands:

cp ".trader_runner/available_tools_store.json" ".trader_runner/available_tools_store_$(date +"%d-%m-%Y")".json
cp ".trader_runner/policy_store.json" ".trader_runner/policy_store_$(date +"%d-%m-%Y")".json
cp ".trader_runner/utilized_tools.json" ".trader_runner/utilized_tools_$(date +"%d-%m-%Y")".json

Tool selection

Sometimes, a mech tool might temporarily return invalid results. As a result, the service would end up performing mech calls without being able to use the response. Assuming that this tool has a large reward rate in the policy weights, the service might end up spending a considerable amount of xDAI before adjusting the tool's reward rate, without making any progress. If a tool is temporarily misbehaving, you could use an environment variable in order to exclude it. This environment variable is defined here and can be overriden by setting it anywhere in the run_service.sh script with a new value, e.g.:

IRRELEVANT_TOOLS=["some-misbehaving-tool", "openai-text-davinci-002", "openai-text-davinci-003", "openai-gpt-3.5-turbo", "openai-gpt-4", "stabilityai-stable-diffusion-v1-5", "stabilityai-stable-diffusion-xl-beta-v2-2-2", "stabilityai-stable-diffusion-512-v2-1", "stabilityai-stable-diffusion-768-v2-1"]

Environment variables

You may customize the agent's behaviour by setting these trader-specific environment variables.

Name	Type	Default Value	Description
ON_CHAIN_SERVICE_ID	int	null	The ID of the on-chain service.
OMEN_CREATORS	list	["0x89c5cc945dd550BcFfb72Fe42BfF002429F46Fec"]	The addresses of the market creator(s) that the service will track.
OPENING_MARGIN	int	300	The markets opening before this margin will not be fetched.
LANGUAGES	list	["en_US"]	Filter questions by languages.
SAMPLE_BETS_CLOSING_DAYS	int	10	Sample the bets that are closed within this number of days.
TRADING_STRATEGY	str	kelly_criterion_no_conf	Trading strategy to use.
USE_FALLBACK_STRATEGY	bool	true	Whether to use the fallback strategy.
BET_THRESHOLD	int	100000000000000000	Threshold (wei) for placing a bet. A bet will only be placed if potential_net_profit - BET_THRESHOLD >= 0.
PROMPT_TEMPLATE	str	With the given question "@{question}" and the 'yes' option represented by '@{yes}' and the 'no' option represented by '@{no}', what are the respective probabilities of 'p_yes' and 'p_no' occurring?	The prompt template to use for prompting the mech.
DUST_THRESHOLD	int	10000000000000	Minimum amount (wei) below which a position's redeeming amount will be considered dust.
POLICY_EPSILON	float	0.1	Epsilon value for the e-Greedy policy for the tool selection based on tool accuracy.
DISABLE_TRADING	bool	false	Whether to disable trading.
STOP_TRADING_IF_STAKING_KPI_MET	bool	true	Whether to stop trading if the staking KPI is met.
AGENT_BALANCE_THRESHOLD	int	10000000000000000	Balance threshold (wei) below which the agent will stop trading and a refill will be required.
REFILL_CHECK_INTERVAL	int	10	Interval in seconds to check the agent balance, when waiting for a refill.
FILE_HASH_TO_STRATEGIES_JSON	list	[["bafybeihufqu2ra7vud4h6g2nwahx7mvdido7ff6prwnib2tdlc4np7dw24",["bet_amount_per_threshold"]],["bafybeibxfp27rzrfnp7sxq62vwv32pdvrijxi7vzg7ihukkaka3bwzrgae",["kelly_criterion_no_conf"]]]	A list of mapping from ipfs file hash to strategy names.
STRATEGIES_KWARGS	list	[["bet_kelly_fraction",1.0],["floor_balance",500000000000000000],["bet_amount_per_threshold",{"0.0":0,"0.1":0,"0.2":0,"0.3":0,"0.4":0,"0.5":0,"0.6":60000000000000000,"0.7":90000000000000000,"0.8":100000000000000000,"0.9":1000000000000000000,"1.0":10000000000000000000}]]	A list of keyword arguments for the strategies.
USE_SUBGRAPH_FOR_REDEEMING	bool	true	Whether to use the subgraph to check if a position is redeemed.
USE_NEVERMINED	bool	false	Whether to use Nevermined.
SUBSCRIPTION_PARAMS	list	[["base_url", "https://marketplace-api.gnosis.nevermined.app/api/v1/metadata/assets/ddo"],["did", "did:nv:01706149da2f9f3f67cf79ec86c37d63cec87fc148f5633b12bf6695653d5b3c"],["escrow_payment_condition_address", "0x31B2D187d674C9ACBD2b25f6EDce3d2Db2B7f446"],["lock_payment_condition_address", "0x2749DDEd394196835199471027713773736bffF2"],["transfer_nft_condition_address", "0x659fCA7436936e9fe8383831b65B8B442eFc8Ea8"],["token_address", "0x1b5DeaD7309b56ca7663b3301A503e077Be18cba"], ["order_address","0x72201948087aE83f8Eac22cf7A9f2139e4cFA829"], ["nft_amount", "100"], ["payment_token","0x0000000000000000000000000000000000000000"], ["order_address", "0x72201948087aE83f8Eac22cf7A9f2139e4cFA829"],["price", "1000000000000000000"]]	Parameters for the Nevermined subscription.

The rest of the common environment variables are present in the service.yaml, which are customizable too.

Checking agents' health

You may check the health of the agents by querying the /healthcheck endpoint. For example:

curl -sL localhost:8716/healthcheck | jq -C

This will return a JSON output with the following fields:

Field	Type	Criteria
seconds_since_last_transition	float	The number of seconds passed since the last transition in the FSM.
is_tm_healthy	bool	false if more than BLOCKS_STALL_TOLERANCE (60) seconds have passed since the last begin block request received from the tendermint node. true otherwise.
period	int	The number of full cycles completed in the FSM.
reset_pause_duration	int	The number of seconds to wait before starting the next FSM cycle.
rounds	list	The last rounds (upto 25) in the FSM that happened including the current one.
is_transitioning_fast	bool	true if is_tm_healthy is true and seconds_since_last_transition is less than twice the reset_pause_duration. false otherwise.

So, you can usually use is_transitioning_fast as a rule to check if an agent is healthly. To add a more strict check, you can also tune a threshold for the seconds_since_last_transition and rate of change of period, but that will require some monitoring to fine tune it.

Backup and Recovery

When executed for the first time, the run_service.sh script creates a number of Gnosis chain accounts:

• one EOA account will be used as the service owner and agent operator,
• one EOA account will be used for the trading agent, and
• one smart contract account corresponds to a Safe wallet with a single owner (the agent account).

The addresses and private keys of the EOA accounts (plus some additional configuration) are stored within the folder .trader_runner. In order to avoid losing your assets, back up this folder in a safe place, and do not publish or share its contents with unauthorized parties.

You can gain access to the assets of your service as follows:

1. Ensure that your service is stopped by running stop_service.sh.

2. Ensure that you have a hot wallet (e.g., MetaMask) installed and set up in your browser.

3. Import the two EOAs accounts using the private keys. In MetaMask, select "Add account or hardware wallet" → "Import account" → "Select Type: Private Key", and enter the private key of the owner/operator EOA account (located in .trader_runner/operator_pkey.txt):

4. Repeat the same process with the agent EOA account (private key located in .trader_runner/agent_pkey.json).

Now, you have full access through the hot wallet to the EOAs addresses associated to your service and you can transfer their assets to any other address. You can also manage the assets of the service Safe through the DApp https://app.safe.global/, using the address located in the file .trader_runner/service_safe_address.txt.

Terminate your on-chain service

If you wish to terminate your on-chain service (and receive back the staking/bonding funds to your owner/operator address in case your service is staked) execute:

./stop_service.sh
./terminate_on_chain_service.sh

RPC-related Error Messages

When updating the service, you may need to re-run the script if you obtain any of the following error messages:

Error: Service unbonding failed with following error; ChainTimeoutError(Timed out when waiting for transaction to go through)

Error: Component mint failed with following error; ChainTimeoutError(Timed out when waiting for transaction to go through)

Error: Service activation failed with following error; ChainTimeoutError(Timed out when waiting for transaction to go through)

Error: Service deployment failed with following error; ChainTimeoutError(Timed out when waiting for transaction to go through)

Error: Service terminatation failed with following error; ChainInteractionError({'code': -32010, 'message': 'AlreadyKnown'})

Build deployments without executing the service

The script builds both a Docker Compose deployment (on ./trader/trader_service/abci_build) and a Kubernetes deployment (on ./trader/trader_service/abci_build_k8s). Then, by default, the script will launch the local Docker Compose deployment. If you just want to build the deployment without executing the service (for example, if you are deploying to a custom Kubernetes cluster), then execute the script as

    ./run_service.sh --build-only