Please enable JavaScript.

Coggle requires JavaScript to display documents.

Academic Literature - Coggle Diagram

- - - - roles
      - goals
    - - Step
      - Task
      - Operation
        
        Memory
        
        history data stored into the memory
        at this operation
        
        Decision
        
        decision-making and action planning
        
        Environment(Tool)
        
        Interactions toward the external system, including other agents and the environment
- - - - tape.metadata.parent_id
        
        the ID of the parent tape of which the current tape is a continuation
        (when applicable)
      - tape.metadata.author
        
        which agent or environment made this tape; either by authoring it, or by
        adding steps to it, or by making a revision of another tape.
      - step.metadata.agent
        
        the hierarchical name of the agent that generated the step.
      - step.metadata.node
        
        the name of the node that generated the step.
      - step.metadata.prompt_id
        
        the identifier (id) of the prompt that led to the generation of this step,
        see the explanation below.
    - - thoughts step: the agent’s inner reasoning steps.
      - observations step : that the agent receives from its environment in response to the
        agent’s actions(step).
      - action step: request external inputs.
    - - Reasoning Node
      - LLM Function Node
      - Goal-Setting Node
      - Reflection Node
      - Acting Node
      - Planning Node
    - - Tape
        
        In TapeAgents we often define a tape type by specifying what specific actions, thoughts, and
        observations classes it can contain,
      - Node
        
        In TapeAgents, one builds the agent from nodes: the basic atoms of intelligence. A node describes one LLM call and the classical symbolic processing of the call’s output.
        
        When an agent runs a node, the node generates a unique ID for the prompt that it builds at this iteration.
        The prompt ID thus serves as the unique identifier of a node execution, i.e., of a specific iteration when the
        node was active. The ID also links the step to the LLM call from the node run so we can trace the origin of
        each step down to the specific prompt and LLM output.
      - Step
        
        Nodes generate new tape entries that we call steps: basic atoms of the agent’s memory. Examples of what an agent can do in a step include making
        a long-term plan, reasoning about how to fulfill the plan or how to use a tool, requesting a tool call.
- - - - Agent User
      - Agent Developer
      - Developer(scaffolding or component foundation models)
      - Tools and services providers
      - Any humans with whom the agent has interacted
    - - Agent’s goal
      - Tools or services
      - Agent’s permissions
      - Sector, e.g., Finance
    - - results of evaluations
      - previous incidents
      - agent dependencies