This repo contains examples of how to use the Fixie.ai platform. You can clone this repository directly to get started.
Built a cool agent that you want to share? Contribute back to this repo by opening a PR!
Full docs can be found here: https://docs.fixie.ai/.
You'll need to install the Fixie CLI to run these examples. You'll need Python >= 3.9 installed.
Install the Fixie CLI with pip install fixieai
. Once installed, run fixie auth
to authenticate to the Fixie service. If you don't already have an account, you'll need to create one at app.fixie.ai.
You can scaffold out a default agent by running: fixie init
Fixie makes it easy to build, test, and debug your agents locally. To build an agent you need two files:
agent.yaml
main.py
In the agent.yaml
file you will have the following fields:
handle: "agent"
description: |
The `agent` serves the following function.
Example queries:
-Do this task for me.
more_info_url: "https://github.com/fixie-ai/fixie-examples"
entry_point: main:agent
public: false
In the main.py
file you will have the following fields:
from fixieai import agents
BASE_PROMPT = """I am an intelligent agent that does a task."""
FEW_SHOTS = """
Q: Example question here
A: Example of answer here
"""
agent = agents.CodeShotAgent(BASE_PROMPT, FEW_SHOTS)
Deploying agents will automatically upload your agent to the Fixie cloud and start serving it immediately on the platform.
Run fixie agent deploy
to create (or update) the agent and upload your fewshots and functions to Fixie.
Once you run this command you test it locally by running the following command: fixie session new '@username/agentname example query
.
Or you can go to app.fixie.ai and chat with your new agent by invoking it in the main chat with @username/agentname
.
We recommend that you write the agent handle in small caps. See example below
handle: "dice"
description: |
The `dice` agent rolls virtual dice.
Example queries:
- Roll a d20.
- Roll two dice and blow on them first for good luck
- Roll 3d8
Tags: func
more_info_url: "https://github.com/fixie-ai/fixie-examples"
entry_point: main:agent
public: false