Skip to main content
Potpie provides three specialized AI agents that understand your codebase through knowledge graph analysis and systematic code exploration.

Ask a Question

Use the Ask Agent for exploration and comprehension

Build a Feature

Use the Code Agent for changes and features

Debug an Issue

Use the Debug Agent for issues and errors

​
How to Use Agents

All agents follow the same usage pattern: create a conversation, send messages, get responses. They differ in what they do with your messages.
1

Create Conversation

Choose your agent and create a conversation
response = requests.post(
    "https://production-api.potpie.ai/api/v1/conversations/",
    headers={"Authorization": "Bearer <firebase_token>"},
    json={
        "user_id": "user_123",
        "title": "Your task description",
        "status": "active",
        "project_ids": ["proj_abc123"],
        "agent_ids": ["codebase_qna_agent"]  # or "code_generation_agent" or "debugging_agent"
    }
)
conversation_id = response.json()["conversation_id"]
2

Send Message

All agents use Form data (not JSON) for messages
message_response = requests.post(
    f"https://production-api.potpie.ai/api/v1/conversations/{conversation_id}/message/",
    headers={"Authorization": "Bearer <firebase_token>"},
    data={
        "content": "Your question, request, or bug report"
    },
    params={"stream": False}
)

answer = message_response.json()["response"]
Message endpoint uses Form data (data=), not JSON (json=). Use multipart/form-data content type.
3

Get Response

All agents return the same response format
{
  response: string           // Markdown-formatted answer
  tool_calls: ToolCall[]     // Tools the agent used
  citations: string[]        // Files referenced
}
Perfect for:
  • “How does this feature work?”
  • “Where is X implemented?”
  • “What files handle Y?”
  • “Show me the authentication flow”
Not ideal for:
  • Making code changes (use Code Agent)
  • Debugging specific bugs (use Debug Agent)