In tasks like customer support, consulting, programming, writing, teaching, etc., language agents can reduce human effort and are a possible first step toward artificial general intelligence (AGI). Recent demonstrations of language agents’ potential, including AutoGPT and BabyAGI, have sparked much attention from researchers, developers, and general audiences. 

Even for seasoned developers or researchers, most of those demos or repositories usually are not conducive to customizing, configuring, and deploying latest agents. This restriction results from the proven fact that these demonstrations are often proof-of-concepts that highlight the potential of language agents quite than being more substantial frameworks that could be used to regularly develop and customize language agents. 

Moreover, studies show that the vast majority of these open-source sources cover only a tiny percentage of the essential language agent abilities, equivalent to job decomposition, long-term memory, web navigation, tool usage, and multi-agent communication. Moreover, most (if not all) of the language agent frameworks currently in use rely exclusively on a temporary task description and fully on the power of LLMs to plan and act. Attributable to the high randomness and consistency across different runs, language agents are difficult to change and tweak, and the user experience is poor.

Researchers from AIWaves Inc., Zhejiang University, and ETH Zürich present AGENTS, an open-source language agent library and framework to support LLM-powered language agents. The goal of AGENTS is to make language agent customization, tuning, and deployment as straightforward as possible—even for non-specialists—while yet being easily expandable for programmers and researchers. The library also offers the core capabilities listed below, which mix to make it a versatile platform for language agents: 

Long-short-term memory: AGENTS incorporate the memory components, allowing language agents to routinely update a short-term working memory with a scratchpad and store and retrieve long-term memory using VectorDB and semantic search. Users can resolve whether to present an agent long-term memory, short-term memory, or each by simply filling up a field within the configuration file. 

Web navigation and the usage of tools: The aptitude of autonomous agents to make use of external tools and browse the web is one other crucial characteristic. AGENTS supports a number of widely used external APIs and offers an abstract class that makes it easy for programmers to include other tools. By classifying web search and navigation as specialized APIs, we also make it possible for agents to browse the web and gather information. 

Multiple-agent interaction: AGENTS permit customizable multi-agent systems and single-agent capabilities, which could be useful for specific applications like games, social experiments, software development, etc. The “dynamic scheduling” function in AGENTS is one latest addition for multi-agent communication. Dynamic scheduling allows establishing a controller agent that serves as a “moderator” and chooses which agent to conduct the following motion based on their roles and up to date history as an alternative of scheduling the order for the agents to act with hard-coded rules. The likelihood exists for more flexible and natural communication between several agents when using dynamic scheduling. By defining the controller’s rule within the configuration file using plain language, developers can quickly alter the controller’s behavior. 

Human-agent interaction is supported by AGENTS in each single-agent and multi-agent scenarios, enabling interaction and communication between a number of humans and language agents.

Controllability: Using a symbolic plan, often generally known as standard operating procedures (SOPs), AGENTS offer a revolutionary paradigm for developing controllable agents. An SOP is a graph with several states that describes the assorted circumstances an agent might face while carrying out a task and the foundations for transitioning between the states. An SOP in AGENTS is a painstakingly recorded collection of detailed instructions that specify how an agent or group of agents should perform a selected activity or procedure. This is analogous to SOPs in the true world. An LLM can produce SOPs that the user can alter while personalizing and fine-tuning the agent. After deployment, an agent will function by the instructions and standards set forth for every state and dynamically change its present state in response to interactions with the skin world, people, or other agents. With the arrival of the symbolic plan, it’s now possible to supply fine-grained control over an agent’s behavior, improving its stability and predictability while facilitating tuning and agent optimization.

The team hopes that AGENTS make it easier for researchers to check language agents, developers to create applications utilizing language agents, and non-technical audiences to create and modify unique language agents. 

Take a look at the Paper and Github. All Credit For This Research Goes To the Researchers on This Project. Also, don’t forget to hitch our 30k+ ML SubReddit, 40k+ Facebook Community, Discord Channel, and Email Newsletter, where we share the newest AI research news, cool AI projects, and more.

If you happen to like our work, you’ll love our newsletter..