# Generalization as a Service

The agent framework described in this paper offers far greater potential than contemporary chatbot-style prompt-and-response agents. However, it also requires far more work to create from scratch. Not only do the feedback loops require intensive calibration and testing, the environment in which the agent runs must also be carefully controlled. 

The agent is - at base - a cyberhazard. While it is necessary to grant it terminal/internet access in order to take advantage of its generalising capabilities, this also gives it the capacity to engage in destructive behaviours. Our own test agents deleted themselves or comprehensively destroyed their containers innumerable times. 

For this reason it is recommended to run the agent only in an isolated environment, using previously tested framework code. To facilitate this for ordinary users, we propose an adapted Forge-style workbench site. 

<figure><img src="/files/NZQ47Y8wVXNS9BgbNrSQ" alt=""><figcaption></figcaption></figure>

This will allow users to specify their own metrics (x.com likes, user interactions, money accumulated...) and the tools to which they wish to give the agent access (i.e. the contents of their .env file), as well as any personality details they wish to include.

<figure><img src="/files/m09CcRg0vgQmMz4abMsB" alt=""><figcaption></figcaption></figure>

This can then be deployed securely and work remotely on the user's behalf via API. This framework can then be used for a variety of purposes. In the  example below, for instance, UI blocks are compiled by the user to create a trading agent that will:

a) Formulate over-arching strategies in response to market conditions;

b) Use tools to respond to contingent events in light of these strategies;

c) Learn from past successes and failures to perform better in the future.&#x20;

<figure><img src="/files/QCmMA7mRTvoJ3gyAqVVw" alt=""><figcaption></figcaption></figure>

Here the users choose four elements:

1. The model to be used (this must be local if retraining is to be applied, but remote models can be used if long term memory is to be created via a RAG vector database);
2. The tools that the AI will be given to research and trade - from wallet keys, to APIs, to news search libraries;
3. The prompts to be given to the agent to guide it through these processes - these determine whether it will, for example, take a more speculative or a more conservative attitude in its training.&#x20;
4. The periodic updates and notifications to which the agent will be exposed. These could be market news, stop loss warnings, or simply tweets from popular influencers. Likewise, the agent is given full freedom to decide how to respond to them. It may choose to sell a coin whose value has increased, or it may decide instead to tweet about the token, thus encouraging new market entrants.&#x20;


---

# Agent Instructions: Querying This Documentation

If you need additional information that is not directly available in this page, you can query the documentation dynamically by asking a question.

Perform an HTTP GET request on the current page URL with the `ask` query parameter:

```
GET https://superioragents.gitbook.io/superior-agents-documents/our-solutions-and-features-1/monetization-strategies/generalization-as-a-service.md?ask=<question>
```

The question should be specific, self-contained, and written in natural language.
The response will contain a direct answer to the question and relevant excerpts and sources from the documentation.

Use this mechanism when the answer is not explicitly present in the current page, you need clarification or additional context, or you want to retrieve related documentation sections.