Planned Feature — Coming Soon

🚧 This feature is not yet released. The documentation below describes planned behavior and may change before launch.

Building Your First Agent

A step-by-step guide to creating a production-ready agent from scratch. We'll build a Company Research Agent that takes a company name and returns a structured profile.

What We'll Build

An agent that:

1. Accepts a company name and optional parameters
2. Searches the web for information
3. Visits the company website
4. Returns a structured JSON profile

Step 1: Plan Your Inputs

Start by deciding what data the agent needs. Think about who will use it:

• If just you — keep inputs minimal
• If marketplace — think about what different users would want to customize

Python

from flymyai import AgentClient

client = AgentClient(api_key="fly-***")

input_schema = {
    "type": "object",
    "properties": {
        "company_name": {
            "type": "string",
            "description": "Company to research"
        },
        "focus": {
            "type": "string",
            "enum": ["general", "financials", "product", "team", "competitors"],
            "default": "general",
            "description": "What aspect to focus on"
        },
        "depth": {
            "type": "string",
            "enum": ["brief", "standard", "thorough"],
            "default": "standard",
            "description": "Level of research depth"
        }
    },
    "required": ["company_name"]
}

JavaScript

import { AgentClient } from "flymyai-js-client";

const client = new AgentClient({ apiKey: "fly-***" });

const inputSchema = {
    type: "object",
    properties: {
        company_name: { type: "string", description: "Company to research" },
        focus: {
            type: "string",
            enum: ["general", "financials", "product", "team", "competitors"],
            default: "general",
            description: "What aspect to focus on"
        },
        depth: {
            type: "string",
            enum: ["brief", "standard", "thorough"],
            default: "standard",
            description: "Level of research depth"
        }
    },
    required: ["company_name"]
};

Step 2: Write the Goal with Templating

Use Handlebars to inject inputs into the goal. Think of it as building a dynamic prompt:

goal = """Research {{ company_name }} and create a comprehensive company profile.

Research focus: {{ focus }}.
Depth level: {{ depth }}.

{{#if focus}}
{{#if (eq focus "financials")}}
Prioritize: revenue, funding rounds, valuation, growth metrics, and financial health indicators.
{{/if}}
{{#if (eq focus "product")}}
Prioritize: core products, pricing, technology stack, unique features, and product roadmap.
{{/if}}
{{#if (eq focus "competitors")}}
Prioritize: identifying top 5 competitors, comparative strengths/weaknesses, market positioning.
{{/if}}
{{/if}}

Return a structured profile with all findings, clearly sourced."""

Step 3: Define Output Schema

Structured output makes the agent predictable and integrable:

output_schema = {
    "type": "object",
    "properties": {
        "company": {
            "type": "object",
            "properties": {
                "name": {"type": "string"},
                "website": {"type": "string"},
                "industry": {"type": "string"},
                "founded": {"type": "string"},
                "headquarters": {"type": "string"},
                "employee_count": {"type": "string"},
                "description": {"type": "string"}
            }
        },
        "key_findings": {
            "type": "array",
            "items": {
                "type": "object",
                "properties": {
                    "category": {"type": "string"},
                    "finding": {"type": "string"},
                    "source": {"type": "string"}
                }
            }
        },
        "summary": {"type": "string"},
        "sources": {
            "type": "array",
            "items": {"type": "string"}
        }
    }
}

Step 4: Create the Agent

Python

agent = client.agents.create(
    name="Company Profiler",
    goal=goal,
    tools=["web_search", "browse_website", "news_search"],
    model="anthropic/default",
    input_schema=input_schema,
    output_schema=output_schema,
    instructions={
        "planning": "1) Search for the company. 2) Visit their website. 3) Check recent news. 4) Compile findings.",
        "tool_calling": "Always visit the company's official website. Cross-reference claims across 2+ sources.",
        "formatting": "Cite sources for every key finding."
    }
)

print(f"Agent created: {agent.id}")

JavaScript

const agent = await client.agents.create({
    name: "Company Profiler",
    goal,
    tools: ["web_search", "browse_website", "news_search"],
    model: "anthropic/default",
    inputSchema,
    outputSchema,
    instructions: {
        planning: "1) Search for the company. 2) Visit their website. 3) Check recent news.",
        toolCalling: "Always visit the official website. Cross-reference across 2+ sources.",
        formatting: "Cite sources for every key finding."
    }
});

Step 5: Test with Different Inputs

Test with minimal input (required fields only):

run1 = client.runs.create(
    agent_id=agent.id,
    input={"company_name": "Stripe"}
)
result1 = client.runs.wait(run1.id)
print(result1.output["summary"])

Test with all fields:

run2 = client.runs.create(
    agent_id=agent.id,
    input={
        "company_name": "Stripe",
        "focus": "financials",
        "depth": "thorough"
    }
)
result2 = client.runs.wait(run2.id)
print(result2.output["key_findings"])

Step 6: Iterate

Review the output. Common improvements:

Issue	Fix
Missing information	Add more specific instructions in planning
Wrong sources	Tighten tool_calling instructions
Poor formatting	Adjust formatting instructions or output schema
Too slow	Reduce depth default, use faster model
Hallucinations	Add "Only include verified, sourced claims" to instructions

client.agents.update(
    agent_id=agent.id,
    instructions={
        "planning": "Updated planning instructions...",
        "tool_calling": "Only cite information from pages you actually visited.",
        "formatting": "Every finding MUST have a source URL."
    }
)

Step 7: Publish (Optional)

Happy with the results? Publish for production use or marketplace:

# For personal production use
client.agents.publish(agent_id=agent.id)

# Or for marketplace
client.agents.update(
    agent_id=agent.id,
    visibility="public",
    pricing={"per_run": 0.50}
)
client.agents.publish(agent_id=agent.id)

Complete Code

Python

from flymyai import AgentClient

client = AgentClient(api_key="fly-***")

# Create agent
agent = client.agents.create(
    name="Company Profiler",
    goal="""Research {{ company_name }} and create a comprehensive company profile.
    Research focus: {{ focus }}. Depth: {{ depth }}.
    Return a structured profile with all findings, clearly sourced.""",
    tools=["web_search", "browse_website", "news_search"],
    input_schema={
        "type": "object",
        "properties": {
            "company_name": {"type": "string", "description": "Company to research"},
            "focus": {
                "type": "string",
                "enum": ["general", "financials", "product", "team", "competitors"],
                "default": "general"
            },
            "depth": {
                "type": "string",
                "enum": ["brief", "standard", "thorough"],
                "default": "standard"
            }
        },
        "required": ["company_name"]
    },
    output_schema={
        "type": "object",
        "properties": {
            "company": {"type": "object", "properties": {
                "name": {"type": "string"},
                "website": {"type": "string"},
                "industry": {"type": "string"},
                "description": {"type": "string"}
            }},
            "key_findings": {"type": "array", "items": {"type": "object", "properties": {
                "category": {"type": "string"},
                "finding": {"type": "string"},
                "source": {"type": "string"}
            }}},
            "summary": {"type": "string"},
            "sources": {"type": "array", "items": {"type": "string"}}
        }
    }
)

# Run
run = client.runs.create(agent_id=agent.id, input={"company_name": "Stripe"})
result = client.runs.wait(run.id)
print(result.output)

JavaScript

import { AgentClient } from "flymyai-js-client";

const client = new AgentClient({ apiKey: "fly-***" });

const agent = await client.agents.create({
    name: "Company Profiler",
    goal: `Research {{ company_name }} and create a comprehensive company profile.
    Focus: {{ focus }}. Depth: {{ depth }}.
    Return a structured profile with all findings, clearly sourced.`,
    tools: ["web_search", "browse_website", "news_search"],
    inputSchema: {
        type: "object",
        properties: {
            company_name: { type: "string", description: "Company to research" },
            focus: { type: "string", enum: ["general", "financials", "product", "team", "competitors"], default: "general" },
            depth: { type: "string", enum: ["brief", "standard", "thorough"], default: "standard" }
        },
        required: ["company_name"]
    },
    outputSchema: {
        type: "object",
        properties: {
            company: { type: "object", properties: { name: { type: "string" }, website: { type: "string" } } },
            key_findings: { type: "array", items: { type: "object" } },
            summary: { type: "string" },
            sources: { type: "array", items: { type: "string" } }
        }
    }
});

const run = await client.runs.create({ agentId: agent.id, input: { company_name: "Stripe" } });
const result = await client.runs.wait(run.id);
console.log(result.output);

Next Steps

• Publishing to Marketplace — monetize your agent
• Multi-Agent Handoffs — build agent teams
• Custom Tools & MCP — extend capabilities