Skip to content

Standard Agent – Getting Started

Standard Agent is a modular framework for building AI agents that can plan, act, and recover from failures. It ships with a ready‑to‑use ReWOO reasoning stack and the Jentic tool platform out of the box—yet every layer is swappable.


Why Standard Agent?

  • 🧩 Modular architecture — swap LLMs, reasoning strategies, tool providers, memory backends
  • 🔄 Autonomous recovery — built‑in detection + intelligent self‑healing
  • ⚡ Ready to ship — ships with ReWOO reasoning + Jentic tools
  • 🔧 Fully extensible — customize components for your stack

New to Business Agents?

Start with the concept overview: Business Agents.


Prerequisites


1) Get a Jentic Agent API key

  1. Sign up: app.jentic.com
  2. In Credentials, add any provider secrets you’ll use (e.g., Discord).
  3. Go to Agents → New Agent, select allowed APIs/workflows, then Generate API key.

Set it in your environment:

\=== "macOS / Linux" bash export JENTIC_AGENT_API_KEY="<your-agent-api-key>" \=== "Windows (PowerShell)" powershell setx JENTIC_AGENT_API_KEY "<your-agent-api-key>" $env:JENTIC_AGENT_API_KEY = "<your-agent-api-key>" # current session

Keys are scoped: each agent only sees what you enable.


2) Install and set up

git clone https://github.com/jentic/standard-agent.git
cd standard-agent

# Install deps (uses the repo's Makefile)
make install

# Activate venv (POSIX shells)
source .venv/bin/activate

Create or update your .env:

# .env
# Jentic (scoped) Agent key
JENTIC_AGENT_API_KEY="your-agent-api-key-here"

# LLM provider (pick one)
OPENAI_API_KEY="your-openai-api-key-here"
ANTHROPIC_API_KEY="your-anthropic-api-key-here"
GEMINI_API_KEY="your-google-gemini-api-key-here"

# Optional default model
LLM_MODEL="claude-sonnet-4"

3) First run (hello, agent)

Create test_agent.py:

import os
from dotenv import load_dotenv
from agents.prebuilt import ReWOOAgent

load_dotenv()

agent = ReWOOAgent(model=os.getenv("LLM_MODEL", "claude-sonnet-4"))
result = agent.solve("What's the weather like in San Francisco?")
print(result)

Run it:

python test_agent.py

You’ll see the agent plan, execute API calls via Jentic, and return a concise result.


4) Interactive mode

Create main.py:

import os
from dotenv import load_dotenv
from agents.prebuilt import ReWOOAgent
from utils.cli import read_user_goal, print_result

load_dotenv()
agent = ReWOOAgent(model=os.getenv("LLM_MODEL", "claude-sonnet-4"))

print("🤖 Agent ready. Ctrl+C to exit.")
while True:
    try:
        goal = read_user_goal()
        if not goal:
            continue
        result = agent.solve(goal)
        print_result(result)
    except KeyboardInterrupt:
        print("\n👋 Bye!")
        break

Run and try goals like:

  • “Find the 3 newest Figshare AI articles and summarize them.”
  • “Create a Zendesk ticket then post a Slack notification with the ticket URL.”
  • “Translate this text to Sith.”

These require the corresponding API credentials set in Jentic → Credentials and the agent scoped to those capabilities.


How it works (high level)

  1. Plan: Agent breaks the goal into steps.
  2. Discover: Tools/capabilities are loaded JITT (Just‑in‑Time Tooling) from Jentic.
  3. Execute: Steps call APIs/workflows with validated inputs.
  4. Reflect & recover: Failures trigger retries / alternative plans.
  5. Summarize: Results are compiled and returned.

Troubleshooting

  • 401 / Unauthorized

  • Check JENTIC_AGENT_API_KEY is exported and the agent is scoped to the API you’re calling.

  • Missing credentials

  • Add provider secrets in the Jentic app under Credentials, then retry.

  • Module not found

  • Activate the venv: source .venv/bin/activate (or re‑run make install).

  • Nothing happens / can’t find tools

  • Confirm the agent has access to those APIs/workflows in the Jentic dashboard.

See also: FAQ


Next steps


Migration notes

  • Deprecated: JENTIC_UUID, JENTIC_API_KEY, jentic register
  • Use now: Create keys in Agents and export JENTIC_AGENT_API_KEY
  • Keys are scoped to selected APIs/workflows; create multiple agents for different apps/bots as needed.