Retrieving Traces

The Agent Run Logs API allows you to retrieve detailed execution logs for agent runs, providing insights into the agent’s thought process, tool usage, and execution flow.

Endpoint

GET /v1/organizations/{org_id}/agent/run/{agent_run_id}/logs

Authentication

This endpoint requires API token authentication. Include your token in the Authorization header:

Authorization: Bearer YOUR_API_TOKEN

Parameters

Parameter	Type	Required	Description
org_id	integer	Yes	Your organization ID
agent_run_id	integer	Yes	The ID of the agent run to retrieve logs for
skip	integer	No	Number of logs to skip for pagination (default: 0)
limit	integer	No	Maximum number of logs to return (default: 100, max: 100)

Response Structure

The endpoint returns an AgentRunWithLogsResponse object containing the agent run details and paginated logs:

{
  "id": 12345,
  "organization_id": 67890,
  "status": "completed",
  "created_at": "2024-01-15T10:30:00Z",
  "web_url": "https://app.codegen.com/agent/trace/12345",
  "result": "Task completed successfully",
  "logs": [
    {
      "agent_run_id": 12345,
      "created_at": "2024-01-15T10:30:15Z",
      "tool_name": "ripgrep_search",
      "message_type": "ACTION",
      "thought": "I need to search for the user's function in the codebase",
      "observation": {
        "status": "success",
        "results": ["Found 3 matches..."]
      },
      "tool_input": {
        "query": "function getUserData",
        "file_extensions": [".js", ".ts"]
      },
      "tool_output": {
        "matches": 3,
        "files": ["src/user.js", "src/api.ts"]
      }
    }
  ],
  "total_logs": 25,
  "page": 1,
  "size": 100,
  "pages": 1
}

Agent Run Log Fields

Each log entry in the logs array contains the following fields:

Core Fields

Field	Type	Description
agent_run_id	integer	The ID of the agent run this log belongs to
created_at	string	ISO 8601 timestamp when the log entry was created
message_type	string	The type of log entry (see Log Types below)

Agent Reasoning Fields

Field	Type	Description
thought	string | null	The agent’s internal reasoning or thought process for this step

Tool Execution Fields

Field	Type	Description
tool_name	string | null	Name of the tool being executed (e.g., “ripgrep_search”, “file_write”)
tool_input	object | null	JSON object containing the parameters passed to the tool
tool_output	object | null	JSON object containing the tool’s execution results
observation	object | string | null	The agent’s observation of the tool execution results or other contextual data

Log Types

The message_type field indicates the type of log entry. Here are the possible values:

Plan Agent Types

Type	Description
ACTION	The agent is executing a tool or taking an action
PLAN_EVALUATION	The agent is evaluating or updating its plan
FINAL_ANSWER	The agent is providing its final response or conclusion
ERROR	An error occurred during execution
USER_MESSAGE	A message from the user (e.g., interruptions or additional context)
USER_GITHUB_ISSUE_COMMENT	A comment from a GitHub issue that the agent is processing

PR Agent Types

Type	Description
INITIAL_PR_GENERATION	The agent is generating the initial pull request
DETECT_PR_ERRORS	The agent is detecting errors in a pull request
FIX_PR_ERRORS	The agent is fixing errors found in a pull request
PR_CREATION_FAILED	Pull request creation failed
PR_EVALUATION	The agent is evaluating a pull request

Commit Agent Types

Type	Description
COMMIT_EVALUATION	The agent is evaluating commits

Link Types

Type	Description
AGENT_RUN_LINK	A link to another related agent run

Field Population Patterns

Different log types populate different fields:

ACTION Logs

• Always have: tool_name, tool_input, tool_output
• Often have: thought, observation
• Example: Tool executions like searching code, editing files, creating PRs

PLAN_EVALUATION Logs

• Always have: thought
• May have: observation
• Rarely have: Tool-related fields
• Example: Agent reasoning about next steps

ERROR Logs

• Always have: observation (containing error details)
• May have: tool_name (if error occurred during tool execution)
• Example: Failed tool executions or system errors

FINAL_ANSWER Logs

• Always have: observation (containing the final response)
• May have: thought
• Example: Agent’s final response to the user

Usage Examples

Basic Log Retrieval

import requests

url = "https://api.codegen.com/v1/organizations/67890/agent/run/12345/logs"
headers = {"Authorization": "Bearer YOUR_API_TOKEN"}

response = requests.get(url, headers=headers)
data = response.json()

print(f"Agent run status: {data['status']}")
print(f"Total logs: {data['total_logs']}")

for log in data['logs']:
    print(f"[{log['created_at']}] {log['message_type']}: {log['thought']}")

Filtering by Log Type

# Get only ACTION logs to see tool executions
action_logs = [log for log in data['logs'] if log['message_type'] == 'ACTION']

for log in action_logs:
    print(f"Tool: {log['tool_name']}")
    print(f"Input: {log['tool_input']}")
    print(f"Output: {log['tool_output']}")
    print("---")

Pagination Example

# Get logs in batches of 50
skip = 0
limit = 50
all_logs = []

while True:
    url = f"https://api.codegen.com/v1/organizations/67890/agent/run/12345/logs?skip={skip}&limit={limit}"
    response = requests.get(url, headers=headers)
    data = response.json()

    all_logs.extend(data['logs'])

    if len(data['logs']) < limit:
        break  # No more logs

    skip += limit

print(f"Retrieved {len(all_logs)} total logs")

Debugging Failed Runs

# Find error logs to debug issues
error_logs = [log for log in data['logs'] if log['message_type'] == 'ERROR']

for error_log in error_logs:
    print(f"Error at {error_log['created_at']}: {error_log['observation']}")
    if error_log['tool_name']:
        print(f"Failed tool: {error_log['tool_name']}")

Common Use Cases

1. Building Monitoring Dashboards

Use the logs to create dashboards showing:

• Agent execution progress
• Tool usage patterns
• Error rates and types
• Execution timelines

2. Debugging Agent Behavior

Analyze logs to understand:

• Why an agent made certain decisions
• Where errors occurred in the execution flow
• What tools were used and their results

3. Audit and Compliance

Track agent actions for:

• Code change auditing
• Compliance reporting
• Security monitoring

4. Performance Analysis

Monitor:

• Tool execution times
• Common failure patterns
• Agent reasoning efficiency

Rate Limits

• 60 requests per 60 seconds per API token
• Rate limits are shared across all API endpoints

Error Responses

Status Code	Description
400	Bad Request - Invalid parameters
401	Unauthorized - Invalid or missing API token
403	Forbidden - Insufficient permissions
404	Not Found - Agent run not found
429	Too Many Requests - Rate limit exceeded

Feedback and Support

Since this endpoint is in ALPHA, we’d love your feedback! Please reach out through:

• Community Slack
• GitHub Issues
• Email: [e-mail]

The structure and fields of this API may change as we gather feedback and improve the service. We’ll provide advance notice of any breaking changes.