Skip to content

Commit

Permalink
AI Agent Span Semantic Convention
Browse files Browse the repository at this point in the history
  • Loading branch information
gyliu513 committed Dec 6, 2024
1 parent da02dac commit 7849133
Showing 1 changed file with 187 additions and 0 deletions.
187 changes: 187 additions & 0 deletions docs/ai-agent/ai-agent-spans.md
Original file line number Diff line number Diff line change
@@ -0,0 +1,187 @@
<!--- Hugo front matter used to generate the website version of this page:
linkTitle: Generative AI traces
--->

# Semantic Conventions for AI Agent Spans

**Status**: [Experimental][DocumentStatus]

<!-- Re-generate TOC with `markdown-toc --no-first-h1 -i` -->

<!-- toc -->

- [SPAN KIND](#span-kind)
- [Name](#name)
- [AI Agent attributes](#ai-agent-attributes)
- [Workflow Attributes](#workflow-attributes)
- [Agent Attributes](#agent-attributes)
- [Tools Attributes](#tools-attributes)
- [Task Attributes](#task-attributes)
- [Interaction Attributes](#interaction-attributes)
- [Examples](#examples)
- [Workflow: Market Analysis Pipeline](#workflow-market-analysis-pipeline)
- [Agents Involved](#agents-involved)
- [DataCollectorAgent](#datacollectoragent)
- [DataAnalystAgent](#dataanalystagent)
- [Tasks and Tools](#tasks-and-tools)
- [Task: Data Collection](#task-data-collection)
- [Tool Utilized: WebScraperTool](#tool-utilized-webscrapertool)
- [Task: Data Analysis](#task-data-analysis)
- [Tool Utilized: DataAnalyzerTool](#tool-utilized-dataanalyzertool)

<!-- tocstop -->

## SPAN KIND

Each step in an AI Agent workflow is treated as a span.

**Span kind:** SHOULD be `CLIENT`. It MAY be set to `INTERNAL` on spans representing call in an AI Agent workflow.

## Name

AI Agent spans MUST follow the overall [guidelines for span names](https://github.com/open-telemetry/opentelemetry-specification/tree/v1.39.0/specification/trace/api.md#span). The **span name** SHOULD be `{ai_agent.operation.name} {ai_agent.workflow.system}`.
Semantic conventions for individual AI Agent systems and frameworks MAY specify different span name format.

- ai_agent.operation.name: The operation being performed (e.g., `workflow.kickoff`,`agent.execution`, `task.execution`, `agent.interaction`).
- ai_agent.workflow.system: The type of agent performing the operation (e.g., `CrewAI`, `LangGraph`, `AutoGen`).

## AI Agent attributes

### Workflow Attributes

| Attribute | Type | Description | Example | Requirement Level | Stability |
| ------------------------------ | ------ | --------------------------------------------------------- | --------------------------------- | ----------------- | --- |
| `ai_agent.workflow.name` | string | Name of the workflow. | `Data Processing Pipeline` | Required | ![Experimental](https://img.shields.io/badge/-experimental-blue) |
| `ai_agent.workflow.tasks` | array | List of tasks included in the workflow. | `["Data Collection", "Analysis"]` | Recommended | ![Experimental](https://img.shields.io/badge/-experimental-blue) |
| `ai_agent.workflow.agents` | array | List of agents participating in the workflow. | `["Agent A", "Agent B"]` | Recommended | ![Experimental](https://img.shields.io/badge/-experimental-blue) |
| `ai_agent.workflow.config` | object | Configuration settings for the workflow. | `{"setting1": "value1"}` | Recommended | ![Experimental](https://img.shields.io/badge/-experimental-blue) |
| `ai_agent.workflow.start_time` | string | Timestamp when the workflow started. | `2024-12-06T10:00:00Z` | Required | ![Experimental](https://img.shields.io/badge/-experimental-blue) |
| `ai_agent.workflow.end_time` | string | Timestamp when the workflow ended. | `2024-12-06T11:00:00Z` | Required | ![Experimental](https://img.shields.io/badge/-experimental-blue) |
| `ai_agent.workflow.end_state` | string | Final state of the workflow (e.g., `success`, `failure`). | `success` | Required | ![Experimental](https://img.shields.io/badge/-experimental-blue) |
| `ai_agent.workflow.end_reason` | string | Reason for the workflow's end, if applicable. | `Completed all tasks` | Recommended | ![Experimental](https://img.shields.io/badge/-experimental-blue) |
| `ai_agent.workflow.system` | string | System or environment where the workflow is executed. | `CrewAI`, `LangGraph`, `AutoGen` | Recommended | ![Experimental](https://img.shields.io/badge/-experimental-blue) |


### Agent Attributes

| Attribute | Type | Description | Example | Requirement Level | Stability |
| ------------------------------ | ------ | ------------------------------------------ | -------------------------------- | ----------------- | --- |
| `ai_agent.agent.name` | string | Name of the agent. | `Researcher Bot` | Required | ![Experimental](https://img.shields.io/badge/-experimental-blue) |
| `ai_agent.agent.role` | string | Role assigned to the agent. | `Data Collector` | Required | ![Experimental](https://img.shields.io/badge/-experimental-blue) |
| `ai_agent.agent.backstory` | string | Background story or context for the agent. | `Specializes in web data mining` | Required | ![Experimental](https://img.shields.io/badge/-experimental-blue) |
| `ai_agent.agent.workflow_name` | string | Name of the workflow the agent is part of. | `Data Processing Pipeline` | Required | ![Experimental](https://img.shields.io/badge/-experimental-blue) |
| `ai_agent.agent.model` | string | Underlying model powering the agent. | `gpt-4` | Required | ![Experimental](https://img.shields.io/badge/-experimental-blue) |
| `ai_agent.agent.tools` | array | List of tools available to the agent. | `["Web Scraper", "Analyzer"]` | Recommended | ![Experimental](https://img.shields.io/badge/-experimental-blue) |

### Tools Attributes

| Attribute | Type | Description | Example | Requirement Level | Stability |
| ------------------------ | ------ | -------------------------------------------- | ----------------- | ----------------- | --- |
| `ai_agent.tool.name` | string | Name of the tool utilized by the agent. | `Web Scraper` | Required | ![Experimental](https://img.shields.io/badge/-experimental-blue) |
| `ai_agent.tool.function` | string | Specific function or capability of the tool. | `Data Extraction` | Recommended | ![Experimental](https://img.shields.io/badge/-experimental-blue) |
| `ai_agent.tool.output` | object | Output produced by the tool. | `{"data": [...]}` | Required | ![Experimental](https://img.shields.io/badge/-experimental-blue) |


### Task Attributes

| Attribute | Type | Description | Example | Requirement Level | Stability |
| --------------------------- | ------- | ------------------------------------------------------------------------ | ---------------------------------- | ----------------- | --- |
| `ai_agent.task.name` | string | Name of the task. | `Data Collection` | Required | ![Experimental](https://img.shields.io/badge/-experimental-blue) |
| `ai_agent.task.agent_name` | string | Name of the agent responsible for the task. | `Agent A` | Required | ![Experimental](https://img.shields.io/badge/-experimental-blue) |
| `ai_agent.task.description` | string | Detailed description of the task. | `Collect data from specified URLs` | Required | ![Experimental](https://img.shields.io/badge/-experimental-blue) |
| `ai_agent.task.output` | object | Output generated from the task. | `{"collected_data": [...]}` | Required | ![Experimental](https://img.shields.io/badge/-experimental-blue) |
| `ai_agent.task.priority` | string | Priority level of the task (e.g., `low`, `medium`, `high`). | `high` | Recommended | ![Experimental](https://img.shields.io/badge/-experimental-blue) |
| `ai_agent.task.state` | string | Current state of the task (e.g., `pending`, `in_progress`, `completed`). | `in_progress` | Required | ![Experimental](https://img.shields.io/badge/-experimental-blue) |
| `ai_agent.task.duration` | integer | Duration taken to complete the task in milliseconds. | `120000` | Recommended | ![Experimental](https://img.shields.io/badge/-experimental-blue) |

### Interaction Attributes

| Attribute | Type | Description | Example | Requirement Level | Stability |
| ----------------------------- | ------ | ------------------------------------------------------------------ | ------------------ | ----------------- | --- |
| `ai_agent.interaction.type` | string | Type of interaction (e.g., `message_exchange`, `task_delegation`). | `message_exchange` | Required | ![Experimental](https://img.shields.io/badge/-experimental-blue) |
| `ai_agent.interaction.source` | string | Identifier of the source agent initiating the interaction. | `Agent A` | Required | ![Experimental](https://img.shields.io/badge/-experimental-blue) |
| `ai_agent.interaction.target` | string | Identifier of the target agent receiving the interaction. | `Agent B` | Required | ![Experimental](https://img.shields.io/badge/-experimental-blue) |
| `ai_agent.interaction.status` | string | Outcome of the interaction (e.g., `success`, `failure`). | `success` | Required | ![Experimental](https://img.shields.io/badge/-experimental-blue) |


## Examples

### Workflow: Market Analysis Pipeline

Description: This workflow aims to analyze market trends by collecting and analyzing data from various sources.

- Attributes:
- ai_agent.workflow.name: Market Analysis Pipeline
- ai_agent.workflow.tasks: ["Data Collection", "Data Analysis"]
- ai_agent.workflow.agents: ["DataCollectorAgent", "DataAnalystAgent"]
- ai_agent.workflow.config: {"schedule": "weekly", "priority": "high"}
- ai_agent.workflow.start_time: 2024-12-06T08:00:00Z
- ai_agent.workflow.end_time: 2024-12-06T10:00:00Z
- ai_agent.workflow.end_state: success
- ai_agent.workflow.end_reason: All tasks completed successfully
- ai_agent.workflow.tags: ["Market Analysis", "Automation"]
- ai_agent.workflow.system: Production

### Agents Involved

#### DataCollectorAgent

- Attributes:
- ai_agent.agent.name: DataCollectorAgent
- ai_agent.agent.role: Data Collection
- ai_agent.agent.backstory: Specializes in gathering data from diverse online sources.
- ai_agent.agent.workflow_name: Market Analysis Pipeline
- ai_agent.agent.model: gpt-4
- ai_agent.agent.tools: ["WebScraperTool"]

#### DataAnalystAgent

- Attributes:
- ai_agent.agent.name: DataAnalystAgent
- ai_agent.agent.role: Data Analysis
- ai_agent.agent.backstory: Proficient in interpreting data to extract meaningful insights.
- ai_agent.agent.workflow_name: Market Analysis Pipeline
- ai_agent.agent.model: gpt-4
- ai_agent.agent.tools: ["DataAnalyzerTool"]

### Tasks and Tools

#### Task: Data Collection

- Assigned Agent: DataCollectorAgent

- Attributes:
- ai_agent.task.name: Data Collection
- ai_agent.task.agent_name: DataCollectorAgent
- ai_agent.task.description: Gather market data from specified online sources.
- ai_agent.task.output: {"data_size": "500MB", "records": 10000}
- ai_agent.task.priority: high
- ai_agent.task.state: completed
- ai_agent.task.duration: 3600000

#### Tool Utilized: WebScraperTool

- Attributes:
- ai_agent.tool.name: WebScraperTool
- ai_agent.tool.function: Data Extraction
- ai_agent.tool.output: {"records_extracted": 10000, "duration": "1h"}

#### Task: Data Analysis

- Assigned Agent: DataAnalystAgent
- Attributes:
- ai_agent.task.name: Data Analysis
- ai_agent.task.agent_name: DataAnalystAgent
- ai_agent.task.description: Analyze the collected data to identify market trends.
- ai_agent.task.output: {"insights": ["Trend A", "Trend B"]}
- ai_agent.task.priority: high
- ai_agent.task.state: completed
- ai_agent.task.duration: 5400000

#### Tool Utilized: DataAnalyzerTool

- Attributes:
- ai_agent.tool.name: DataAnalyzerTool
- ai_agent.tool.function: Data Analysis
- ai_agent.tool.output: {"identified_trends": 2, "duration": "1.5h"}
- This simplified workflow demonstrates how two agents collaborate to perform data collection and analysis tasks, utilizing specific tools to achieve the objectives of the Market Analysis Pipeline.

0 comments on commit 7849133

Please sign in to comment.