Skip to main content
Version: v2.0

aiXplain Agent Builder Skill

Build, run, debug, and deploy aiXplain agents using plain English — no IDE, no boilerplate, no guesswork.

Drop this skill into any skill-compatible AI coding agent and start describing what you want. The agent handles the architecture, writes the code, deploys to aiXplain, runs a smoke test, and hands you a working agent ID.

🧩 Download the aiXplain Agent Builder skill

Get the latest packaged skill folder, refreshed with the newest Agent Builder updates, and drop it into your AI coding agent.

Download aixplain-agents.zip

SDK version: aixplain==0.2.43 · Last updated: 2026-04-01

Who it's for​

PersonaWhat they get
Non-technical buildersZero-code agent creation — just describe what you want
DevelopersProduction-ready Python, SDK patterns, OAuth wiring, error handling
Power usersExport any deployed agent back to Python, reverse-engineer team architectures, add inspectors and governance
AI teamsSafe API key handling, pre-build approval gates, no silent duplicates

Zero-setup requirements​

  1. Get an aiXplain API key.
  2. Download the aixplain-agents/ folder.
  3. Drag it into your AI coding agent's chat window, for example Claude Code or Cursor, or point it to the folder path.
  4. Say: "Add this as a skill"
  5. Open the deployed agent in aiXplain Studio to inspect traces and runtime behavior.

That's it. No pip installs, no config files, no environment setup — the skill auto-discovers your API key and walks you through anything it can't find.

What you can build​

Build​

Describe the agent in plain English. The skill drafts the plan, tools, output format, and guardrails before it builds.

"Build an agent that searches the web, finds competitor pricing, and emails me a summary"
"Create a support triage agent connected to our knowledge base and Jira"
"Make a real estate evaluator that looks up location data and computes cap rate"

Run / Debug​

Give it an agent ID or name. The skill runs it, inspects the steps, and diagnoses failures.

"Run agent 69ce064f44eef3c9e3850d95 with this query: Austin TX, 450k asking price"
"Debug why the agent isn't using the knowledge base"
"Inspect the execution steps from the last run"

Deploy​

Once approved, the skill creates or updates the agent, connects the right tools, runs a smoke test, and leaves you with a working asset.

"Deploy this customer support agent to aiXplain and test it with a ticket triage prompt"
"Build this real-estate evaluator and deploy it to my workspace"
"Create the team agent, connect the tools, and validate the first run"

What the skill does for you​

  • Shows a pre-build plan before creating anything
  • Finds API keys from ~/.env, repo .env, or environment variables
  • Reuses existing agents instead of creating duplicates
  • Surfaces OAuth connection links when integrations need them
  • Runs a smoke test and checks for grounded output
~/.env → repo .env → environment variables → asks you

Keys are set as environment variables, never hardcoded into generated files.

Best practices baked in​

  • Uses a strong instruction format: ROLE, CONSTRAINTS, and OUTPUT RULES
  • Keeps tool descriptions short and action-oriented for better selection
  • Picks practical iteration limits for lookup vs. deep research tasks
  • Scopes tool actions to the minimum required set
  • Chooses speed vs. quality settings based on the use case
runResponseGeneration=False + output_format="text"  ->  ~30-40% faster
runResponseGeneration=True  + output_format="markdown"  ->  richest output

Sample queries to try​

Web research agent --- "Build an agent that searches the web and summarizes findings into a markdown report"
Support triage --- "Create a support agent that searches our KB, checks ticket history in Jira, and classifies by severity"
Data analyst --- "Build an agent that runs Python on uploaded CSVs and produces charts"
Email assistant --- "Make an agent that reads my Gmail, groups by topic, and drafts replies"
Deploy --- "Build this agent, deploy it to my workspace, and validate the first run"
Debug --- "Why did agent 69abc… return a hallucinated answer instead of using the KB?"
Knowledge base --- "Create an agent connected to our product docs KB that answers support questions"

Integrations supported​

The skill has access to 600 integrations including:

  • Productivity: Gmail, Google Drive, Google Sheets, Notion, Confluence
  • Dev & project: GitHub, Jira, Linear
  • CRM & sales: HubSpot, Salesforce
  • Communication: Slack
  • Data: PostgreSQL, SQLite, aiR Knowledge Base
  • Compute: Python Sandbox, Code Execution
  • Web: Tavily, Firecrawl, Google Search, Google Places
  • Universal: MCP Server and similar extensible connectors

File structure​

aixplain-agents/
├── SKILL.md                      # Core skill — loaded automatically
└── references/
    ├── asset-ids.md              # Curated model, tool, and integration IDs
    ├── integration-playbooks.md  # Connection patterns per integration
    └── inspector-analytics.md    # Governance policies and inspector patterns

Reference files are loaded on demand — only when the task requires them. This keeps context lean and responses fast.