power-userssubagentsautomationtutorialagentic-ai

Building Custom Subagents in aiFetchly: Power User Guide

September 25, 2026

aiFetchly Subagent Management console showing configured specialist agents, runtime limits, and output schemas

Why Monolithic AI Chats Fail Complex Workflows

Power users running high-frequency market research and automated outreach quickly hit the limitations of monolithic AI chats. When a single conversational agent handles search retrieval, entity extraction, data verification, and email drafting in a continuous session, context windows saturate and reasoning degrades. Furthermore, granting broad tool permissions to a general-purpose assistant risks runaway execution loops and unauthorized actions.

AI Fetchly solves this architectural challenge with AI Subagents. Rather than overloading the primary assistant, aiFetchly delegates targeted jobs to scoped specialists. Each subagent operates within isolated context boundaries, governed by explicit tool allowlists, strict runtime budgets, and validated JSON output schemas.

Subagent Roles and Execution Lifecycle

Subagents are not invoked directly by manual prompt commands. Instead, aiFetchly injects an inventory of registered agents into the primary AI context. When a task requires specialized execution, the main orchestrator calls the run_subagent tool with the agent’s unique ID slug.

Subagent Mode	Operational Role	Typical Tool Access	Primary Deliverable
Specialist	Deep domain extraction and retrieval	`google_search`, `knowledge_library_search`	Verified factual profiles
Verifier	Independent validation of claims	Local file inspection, schema validators	Confidence scoring and anomaly alerts
Coordinator	Multi-phase task planning and decomposition	Workspace status tools, dependency checks	Execution roadmaps
Formatter	Reshaping raw outputs into business formats	Structured template transformers	Validated JSON / Markdown tables

Architecture diagram of aiFetchly subagent dispatch pipeline showing main AI caller, run_subagent tool contract, tool isolation, and structured JSON return

Step-by-Step Tutorial: Creating a Custom Specialist Subagent

Follow this walkthrough to build and deploy a dedicated market intelligence subagent in aiFetchly.

Step 1: Open the Subagent Manager

Launch aiFetchly on your Windows desktop.
Click System Setting in the left navigation menu.
Select Manage Subagents to open the subagent catalog.
Click Add Subagent in the upper toolbar.

Step 2: Configure Agent Metadata and Functional Mode

Name: Enter Market Intelligence Researcher.
ID Slug: aiFetchly automatically generates agent-market-intelligence-researcher (locked upon creation).
Description: Enter Gathers company background, business model, and recent hiring signals for prospect domains.
Mode: Select specialist from the role dropdown.

Step 3: Define System Prompt and Tool Allowlists

Subagents require completely self-contained instructions. Because subagents do not inherit parent chat histories, provide explicit instructions:

You are a specialist market researcher. Analyze the target company URL.
Identify: (1) Core value proposition, (2) ICP profile, (3) Pricing model.
Cite every factual finding with a verified source URL.
Return your findings strictly formatted according to the configured JSON schema.

In the Allowed Tools field, enter only the necessary primitives: google_search, scrape_urls_from_search_engine, knowledge_library_search.

aiFetchly strictly intersects this allowlist with active security policies. Subagents cannot send emails or mutate workspace databases.

Step 4: Enforce Safety Limits and Output Schemas

Under runtime parameters, define defensive operational bounds:

Max Tool Calls: Set to 10 to stop recursive retrieval loops.
Max Runtime (Seconds): Set to 120 seconds.
Max Continue Calls: Set to 3 turns.
Output Schema: Define a JSON schema requiring companyName, productsServices, sourceUrls, and confidenceScore.

Click Save. Your subagent is immediately available to the main AI in AI Chat V2 and Plan Mode.

Scalable Desktop Intelligence with Bounded Execution

By delegating intensive extraction to isolated, schema-enforced subagents, power users eliminate context degradation while preventing accidental tool abuse. Explore the full specification in the AI Fetchly Docs.

Visual Generation Prompts

Hero Image Prompt: Modern software interface of aiFetchly Subagent Manager in dark mode, showcasing active specialist agents, tool budget cards (max tool calls, runtime limits), and clean glowing status badges, isometric 3D perspective, electric cyan and emerald green accents.
Body Image Prompt: Technical system architecture diagram illustrating the four-stage subagent dispatch lifecycle: main AI orchestrator, run_subagent contract, isolated sandbox with tool allowlist guardrails, and validated JSON schema return, vector blueprint style.