Agent-E SOP

Appearance

Test ​

Status: NEEDS-ACCOUNT (GitHub clone + API key required) ​

Classification: NEEDS-ACCOUNT

Agent-E is not available as a published pip package.

  • pip show agent-e returns: not found

Agent-E is a GitHub-only Python project at github.com/EmergenceAI/Agent-E. It requires a manual clone, pip install from the repository, Playwright browser setup, and a valid LLM API key (OpenAI or compatible) before it can be used.

What is required ​

  1. Clone the repository:

    bash
    git clone https://github.com/EmergenceAI/Agent-E.git
cd Agent-E

  2. Install dependencies:

    bash
    pip install -e .
playwright install chromium

  3. Set your LLM API key in the environment or .env file (see repo README for exact variable names).

Smoke test: run the built-in test suite (after manual install) ​

Agent-E ships a test runner that executes tasks against real websites. Run a single task (index 0) to verify the environment is wired up correctly:

bash
python -m test.run_tests --min_task_index 0 --max_task_index 1

What success looks like ​

The test runner launches a browser session, sends the first benchmark task to the agent, and prints the agent's step-by-step plan and actions to stdout. A passing run produces output similar to:

Running task 0: <task description>
[Agent] Planning steps...
[Agent] Step 1: Navigate to <url>
[Agent] Step 2: <action>
...
[Agent] Task completed.
Result: PASS

The exact output varies because Agent-E operates on live websites that change over time. What matters is that the agent launches, connects to the browser, calls the LLM, and executes at least one browser action without crashing.

Important notes on live-site tests ​

  • Tests run against real, live websites. Results vary based on current site layout and content.
  • A task result of FAIL or UNCERTAIN does not necessarily mean your installation is broken. It may mean the target site changed.
  • Focus on whether the agent starts, connects to the browser, and makes LLM calls. Those signals confirm the environment is functional.

The fastest way to verify the install without running the full test suite:

  1. Start Agent-E:

    bash
    python -m ae.main

  2. When the browser opens and the chat icon appears, type:

    go to google.com and tell me what you see on the page

  3. Success: Agent-E navigates to Google, reads the page, and returns a description. The terminal shows the LLM plan and step-by-step browser actions with no Python tracebacks.

Verifying the FastAPI server ​

bash
# Terminal 1: start the server
uvicorn ae.server.api_routes:app --loop asyncio

# Terminal 2: send a test task
curl --location 'http://127.0.0.1:8000/execute_task' \
--header 'Content-Type: application/json' \
--data '{"command": "go to google.com and report the page title"}'

Success: the curl response streams back the agent steps and returns the page title (Google) with no HTTP errors.

Notes ​

  • tested=false: there is no pip package and no one-liner install. All tests require a manual clone, Playwright setup, and a valid LLM API key. Commands are documented from the repository source.