Oct 25 • General discussion
Getting Started with AgentKit
AgentKit is a framework/toolkit for building, orchestrating, testing and deploying AI-agents (from single-model agents to multi-agent systems).
agentkit.inngest.com
+2
agentkit.inngest.com
+2
Key features:
Agents + Networks + Routers + Tools + State.
agentkit.inngest.com
+1
Visual workflow builder (“Agent Builder”) + Chat embedding (“ChatKit”) + external connectors/tools.
OpenAI
+1
Built for production-scale orchestration (not only simple prompts).
blott.com
+1
Because you’re already building automated workflows (you use platforms like Make.com + Avatar generation + publishing pipelines), AgentKit can nicely plug into your stack (e.g., agents that drive content workflows, route tasks, integrate tools).
2. Prerequisites & Setup
Before diving in, ensure you have the basics in place:
Access to an LLM provider (e.g., OpenAI API key) or a supported model.
Skywork
+1
Language/Stack knowledge: JS/TypeScript (Node.js) is supported.
agentkit.inngest.com
+1
Basic familiarity with tools, workflows, tool-calls (you already have this via your automation experience).
Strategic thinking about the “agent task” you want: what job it will do, which tools it needs, what scope.
(Optional but beneficial) Familiarity with workflow automation concepts (you have Make.com, Google Sheets, asset generation pipelines) — you can treat AgentKit as a “workflow engine” for AI-agents.
3. Quick-Start Example
Here’s how you can spin up a minimal AgentKit agent locally — you’ll get a “hello agent” and then you can build on it.
For example, using the Inngest version:
Create a JS/TS project:
mkdir my-agentkit-project && cd my-agentkit-project
npm init -y
npm install @inngest/agent-kit inngest
agentkit.inngest.com
2. Create an index.ts (or .js) file:
import { createAgent, anthropic } from '@inngest/agent-kit';
const dbAgent = createAgent({
name: 'Database administrator',
description: 'Expert at PostgreSQL schema/indexes/extensions',
system: 'You are a PostgreSQL expert. You only answer questions about PostgreSQL schema, indexes, extensions.',
model: anthropic({ model: 'claude-3-5-haiku-latest' })
});
// Later you’ll hook this into a server/endpoint to call it.
agentkit.inngest.com
3. Run a local dev server, invoke your agent via HTTP or CLI (per quick-start guide).
agentkit.inngest.com
4. Once the “hello agent” works (you can send an input, get an output), you expand: add a tool that the agent can call (e.g., database query tool), then build a Network (multiple agents collaborating), then add routing logic, logs/traces, etc.
For example, using the OpenAI AgentKit (platform version):
Log into OpenAI’s Agent Builder.
Composio
+1
Pick a template or blank canvas, define a Start node → follow through building the workflow visually.
Composio
Connect tools, define guardrails, publish/embed ChatKit.
This path might fit if you want less code and more visual orchestration.
4. Map AgentKit to Your Use-Case
Given your background (automated workflows for content creators, YouTube pipeline, tools, thumbnails, etc.), here’s how you could apply AgentKit:
Build an Agent that takes a “video asset generation” task: input = YouTube video idea; tool calls = OpenAI prompt generation, thumbnail generator, Google Sheet API, etc.
Build a Network of agents: one agent writes script, another generates thumbnail, another schedules upload via YouTube API & updates Google Sheet.
Use Routers/State in AgentKit to decide which agent handles what (e.g., if the idea is “how to use ElevenLabs voice” route to Script-Agent; if “thumbnail style” route to Thumbnail-Agent).
Embed the agent via UI (using ChatKit) so your users (content creators) can trigger tasks via chat or button.
Add logging/tracing so you can analyze agent performance and retention triggers (fits your expertise in retention automation).
This aligns well with “automated workflows” you already do via Make.com. AgentKit adds the “intelligent agent orchestration layer”.
5. Best Practices & Pitfalls
Here are some tips to keep your first build smooth:
Start small: Build a simple agent with narrow scope (read-only tool) before giving it full autonomous powers.
Skywork
Clear system instructions: Give the agent a crisp “You are …” description so it stays focused.
Tool-permission control: Only allow tool access when you’ve tested.
Add guardrails: If your agent triggers actions (uploading, publishing) you may want human approval or log review.
Iterate with evaluation: Use small datasets of representative inputs; trace outputs and refine.
Skywork
+1
Version everything: Because you’ll likely iterate workflows fast, keep versions of your agent definitions, schemata, etc.
Integration with your stack: Since you have a Google Sheet, Make.com, YouTube pipeline, think about how the agent will hook into your existing APIs. AgentKit may talk to external tools — ensure auth, security, and API access are handled.
Monitor cost/usage: LLM calls have cost; agents that call tools/systems may incur further cost.
Production vs prototype: For your initial builds, treat it as a prototype. Later you can scale, add monitoring, error-handling, observability.
6. Next Steps for YOU
Given your goals (helping creators with automated workflows, faceless YouTube automation, etc), here’s a recommended sequence:
Pick a pilot use-case: e.g., an agent that takes “YouTube topic idea” → generates script + thumbnail prompt → writes to Google Sheet.
Set up a minimal agent in AgentKit following the quick-start above (choose TypeScript/Node).
Integrate one tool your stack already uses (e.g., Google Sheets API) so agent writes metadata to sheet.
Test the agent with real inputs; review outputs manually.
Once stable, expand: add second agent (thumbnail generation), routing logic between agents, publish trigger via YouTube API or Make.com trigger.
Embed a UI: Use ChatKit or your own minimal UI so a creator can trigger via button/interaction.
Build evaluation metrics: track success rates, errors, feedback loops; iterate.
After internal success, package the agent workflow as a service/offering you can sell to your creator target audience.
2
8 comments
Getting Started with AgentKit
powered by
skool.com/shanes-ai-wealth-collective-1691
Build AI video businesses that run on autopilot. Learn to create content automatically and turn AI tools into profitable income streams.
Suggested communities
Powered by